whatmodel.sh · find the best AI models you can run

Best AI models for coding to run locally on a 64GB Mac

A leaderboard of the best AI models for coding you can run locally on a Mac with 64GB of memory. Qwen3.8 27B is #1 right now, followed by Ornith 1.5 35B-A3B and KAT-Coder-V2.5-Dev (35B-A3B). Models are ranked on coding tests like HumanEval, SWE-bench and Aider. Each model shows the download that fits and the command to run it.

322 models · scores combine 59 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][agent tasks][coding][chat & writing][math][vision & documents][web search]

Scores use reviewed coding evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 64GB RAM · Apple Silicon → ~48GB fast · local only · reset
+ more
chip[M1 Max][M1 Ultra][M2 Max][M2 Ultra][M3 Max (16-core CPU / 40-core GPU)][M4 Pro][M4 Max (16-core CPU / 40-core GPU)][M5 Pro][M5 Max (40-core GPU)]
set[all sets][Arena WebDev (Code Arena) · 549][LiveCodeBench v5 (window unspecified) · 7][HumanEval · 6][BigCodeBench · 4][BigCodeBench · 4][HumanEval · 3][LiveCodeBench (version unspecified) · 3][LiveCodeBench 2024-08–2025-05 (version unstated) · 2][HumanEval · 2][HumanEval+ · 2][LiveCodeBench 2024-08–2025-05 (version unstated) · 2][HumanEval · 2][HumanEval · 2][Tulu 2: CodexEval pass@10 · 2][HumanEval · 2]

Filter the model list while keeping task scores. Models without a shared score scale are not yet calibrated; select a set to see its own scores.

Detect your exact chip or graphics card:

curl -G whatmodel.sh/ -d max=0 --data-urlencode "mac=$(sysctl -n hw.memsize machdep.cpu.brand_string)"Mac

curl -G whatmodel.sh/ -d max=0 --data-urlencode "ram=$(awk '/MemT/{print $2}' /proc/meminfo)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null)"Linux

curl.exe -G whatmodel.sh/ -d max=0 --data-urlencode "ram=$((gcim Win32_ComputerSystem).TotalPhysicalMemory)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null)"Windows

$ terminal

detecting your OS…

Run curl "whatmodel.sh/?ram=64&gpu=apple&max=0"to get this list in your terminal (add &format=json for agents).

ram[8][16][24][32][48][64][96][128][192][256]GB
gpu[apple silicon][graphics card][none]
price[local][$0.02][$0.1][$0.5][$2][$10][any]max per M

detecting your OS…

Score vs memory to run

#modeltask score↓where↓speed↓RAM↓context↓get itsources
#modeltask score↓where↓speed↓RAM↓context↓get itsources
31Qwen3.8 27B · 8-bit Alibaba75±1local~36GB262kOllama › LMArena leaderboard datasetQwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Qwen/Qwen3.8-27B model cardRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
64Ornith 1.5 35B-A3B · 8-bit DeepReinforce68±3†local~41GB262kllama.cpp › ornith-ai/Ornith-1.5-35B-A3B model card
69KAT-Coder-V2.5-Dev (35B-A3B) · 8-bit Kuaishou (Kwaipilot)67±3†local~40GB262kllama.cpp › Kwaipilot/KAT-Coder-V2.5-Dev model card
84Qwen3.6 27B · 8-bit Alibaba66±2local~37GB262kOllama › Qwen/Qwen3.6-27B model cardQwen/Qwen3.8-27B model cardmeta-models/Muse-Glimmer-30B model card (speed and benchmark tables)Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
98Qwen3.6 35B-A3B · 4-bit Alibaba64±1local~41GB262kLM Studio › Qwen/Qwen3.6-27B model cardKwaipilot/KAT-Coder-V2.5-Dev model cardQwen/Qwen3.6-35B-A3B model cardRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
108Nemotron Cascade 2 30B-A3B · 8-bit NVIDIA63±2†local~37GB256kllama.cpp › NVIDIA — Nemotron-Cascade 2 technical report
110Qwen3.5 27B · 8-bit Alibaba63±1local~37GB262kOllama › LMArena leaderboard datasetQwen/Qwen3.6-27B model cardKwaipilot/KAT-Coder-V2.5-Dev model cardQwen/Qwen3.6-35B-A3B model cardQwen/Qwen3.5-27B model cardQwen/Qwen3.5-122B-A10B model card
113Ornith 1.0 35B (35B-A3B) · 4-bit DeepReinforce63±2local~41GB262kLM Studio › Kwaipilot/KAT-Coder-V2.5-Dev model cardornith-ai/Ornith-1.5-35B-A3B model card
116Muse Glimmer 30B · 8-bit Meta63±1local~40GB131kOllama › LMArena leaderboard datasetmeta-models/Muse-Glimmer-30B model card (speed and benchmark tables)Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau LeetcodeQwen/Qwen3.8-27B model card
118Laguna XS 2.1 (33B-A3B) · 8-bit Poolside63±3†local~39GB262kOllama › Poolside vendor-reported benchmark scores for its own Laguna models (Hugging Face model cards)
119Gemma 4 31B · 4-bit Google62±1local~37GB262kLM Studio › LMArena leaderboard datasetKwaipilot/KAT-Coder-V2.5-Dev model cardmeta-models/Muse-Glimmer-30B model card (speed and benchmark tables)Gemma Team — Gemma 4 Technical ReportRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau LeetcodeQwen/Qwen3.6-27B model card+1
133Gemma 4 26B-A4B · 4-bit Google61±1local~31GB262kLM Studio › LMArena leaderboard datasetKwaipilot/KAT-Coder-V2.5-Dev model cardGemma Team — Gemma 4 Technical ReportRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcodegoogle/diffusiongemma-26B-A4B-it model cardQwen/Qwen3.6-35B-A3B model card
135MiMo-V2.6-Distill-Qwen-9B · 8-bit Xiaomi61±3*†local~12GBllama.cpp › XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B model card
136Seed-OSS 36B Instruct · 8-bit ByteDance Seed61±3†local~48GB524kllama.cpp › ByteDance-Seed/Seed-OSS-36B-Instruct model card
141Nemotron 3.5 Lightning 30B-A3B · 8-bit NVIDIA60±5*local~42GB262kOllama › Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
142Gemma 4 12B · 8-bit Google60±3local~15GB262kllama.cpp › Gemma Team — Gemma 4 Technical ReportXHToken/Spark-X2.5-4B model card
143Laguna XS.2 · 4-bit Poolside60±1local~39GB262kLM Studio › LMArena leaderboard datasetPoolside vendor-reported benchmark scores for its own Laguna models (Hugging Face model cards)Poolside — Laguna M.1/XS.2 Technical Report
144Devstral Small 2 24B · 8-bit Mistral59±4*†local~32GB262kOllama › mistralai/Devstral-Small-2-24B-Instruct-2512 model card
146Nemotron-Labs-3 Puzzle 75B-A9B · 2-bit NVIDIA59±4*†local~35GBllama.cpp › NVIDIA vendor-reported benchmark scores for its own Nemotron models (Hugging Face model cards)
153Spark-X2.5 4B · 16-bit iFLYTEK (SparkLLM)59±2†local~10GB1MOllama › XHToken/Spark-X2.5-4B model card
159Qwen3-Next 80B-A3B Thinking · 2-bit Alibaba58±4*†local~27GB131kllama.cpp › Qwen/Qwen3.5-4B model cardQwen/Qwen3-Next-80B-A3B-Thinking model card
160Qwen3.5 35B-A3B · 8-bit Alibaba58±1local~47GB262kOllama › LMArena leaderboard datasetNVIDIA — Nemotron-Cascade 2 technical reportKwaipilot/KAT-Coder-V2.5-Dev model cardQwen/Qwen3.6-35B-A3B model cardQwen/Qwen3.5-27B model cardQwen/Qwen3.5-122B-A10B model card
163Phi-4-reasoning 14B · 8-bit Microsoft58±3*†local~21GB33kOllama › Microsoft — Phi-4-reasoning Technical Reportmicrosoft/Phi-4-reasoning model card
167Granite 4.2 30B · 8-bit IBM58±2†local~39GB131kOllama › ibm-granite/granite-4.2-30b model card
168Ministral 3 8B Reasoning 2512 · 4-bit Mistral58±6*†local~11GB262kLM Studio › mistralai/Ministral-3-8B-Reasoning-2512 model cardmistralai/Ministral-3-3B-Reasoning-2512 model card
169Mistral Medium 3.5 (128B) · 2-bit Mistral57±1local~42GB262kllama.cpp › LMArena leaderboard datasetSolar Open 2 Technical Report (Upstage, arXiv:2607.20062v2)
176MiniCPM5 2B · 4-bit OpenBMB57±2†local~4.3GB131kLM Studio › openbmb/MiniCPM5-2B model card
178QwQ-32B-Preview · 8-bit Alibaba57±4*local~43GB33kOllama › LiveCodeBench official website results (2025-08-01)
185GLM-4.5-Air · 2-bit Z.ai56±3local~35GB131kllama.cpp › GLM-4.5 Team (Zhipu AI & Tsinghua) — GLM-4.5 technical reportPrime Intellect INTELLECT-3 model card: Evaluation table (Environments Hub runs)
189Nemotron-Cascade-14B-Thinking · 8-bit NVIDIA56±1†local~18GBllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
190Mellum2.1 12B-A2.5B Thinking · 8-bit JetBrains56±3†local~15GBllama.cpp › JetBrains/Mellum2.1-12B-A2.5B-Thinking model card
191gpt-oss-20b · 4-bit OpenAI56±2local~25GB131kLM Studio › OpenAI — gpt-oss-120b & gpt-oss-20b Model CardFalcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time ScalingRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau LeetcodeByteDance-Seed/Seed-OSS-36B-Instruct model card
192Qwen3-Next 80B-A3B Instruct · 2-bit Alibaba56±2†local~27GB131kllama.cpp › Qwen/Qwen3-Next-80B-A3B-Instruct model card
197Phi-4-reasoning-plus 14B · 8-bit Microsoft55±2local~18GBllama.cpp › Microsoft — Phi-4-reasoning Technical Reportmicrosoft/Phi-4-reasoning model cardFalcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling
198Qwen3-Coder-Next (80B-A3B) · 2-bit Alibaba55±3*local~27GB262kllama.cpp › Vector Institute Evaluation Leaderboard (Inspect Evals runs)
201Qwen3-VL 32B Thinking · 8-bit Alibaba55±3*†local~39GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
203Apriel 1.6 15B Thinker · 8-bit ServiceNow55±2*†local~21GBllama.cpp › ServiceNow-AI/Apriel-1.6-15b-Thinker model card
204OpenReasoning-Nemotron-32B · 8-bit NVIDIA54±2†local~38GBllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
205OpenThinker 32B · 8-bit Open Thoughts54±9*†local~38GB33kllama.cpp › open-thoughts/OpenThinker-32B model card
207Qwen3-VL 30B-A3B Thinking · 8-bit Alibaba54±3*†local~37GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
208Qwen3.5 9B · 4-bit Alibaba54±1local~12GB262kLM Studio › JetBrains/Mellum2.1-12B-A2.5B-Thinking model cardXiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B model cardJetBrains/Mellum2-12B-A2.5B-Instruct model cardXHToken/Spark-X2.5-4B model cardRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau LeetcodeQwen/Qwen3.5-4B model card+3
209Qwen3 30B-A3B Thinking 2507 · 8-bit Alibaba54±3local~41GB262kOllama › ByteDance-Seed/Seed-OSS-36B-Instruct model cardQwen/Qwen3.5-4B model cardQwen/Qwen3-30B-A3B-Thinking-2507 model card
210Command A Reasoning (08-2025) · 2-bit Cohere54±7*†local~34GBllama.cpp › Cohere vendor-reported benchmark scores for its own models (cohere.com blog)
212Qwen2.5 Coder 32B Instruct · 8-bit Alibaba53±1local~43GB131kOllama › EvalPlus / EvalPerf published evaluation runsAider polyglot benchmark runsAider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)LiveCodeBench official website results (2024-12-26)+2
218Qwen3 32B · 8-bit Alibaba53±1local~43GB131kOllama › Aider polyglot benchmark runsQwen/Qwen3-Next-80B-A3B-Instruct model cardByteDance-Seed/Seed-OSS-36B-Instruct model cardFalcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time ScalingWebApp1K models leaderboard (onekq-ai)Qwen/Qwen3-Next-80B-A3B-Thinking model card
219Granite 4.2 8B · 8-bit IBM53±2†local~11GB131kllama.cpp › ibm-granite/granite-4.2-30b model card
222Nemotron-Cascade-8B · 8-bit NVIDIA53±2†local~11GBllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
223Granite 4.1 8B · 4-bit IBM53±1local~11GB131kLM Studio › LMArena leaderboard datasetibm-granite/granite-4.1-8b model card
224Nemotron-Cascade-8B-Thinking · 8-bit NVIDIA53±2†local~11GBllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
225Ministral 3 3B Reasoning 2512 · 4-bit Mistral52±6*†local~5.3GB262kLM Studio › mistralai/Ministral-3-8B-Reasoning-2512 model cardmistralai/Ministral-3-3B-Reasoning-2512 model card
227Qwen2.5 32B Instruct · 8-bit Alibaba52±1local~43GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)WildEval ZeroEval published results
228Qwen3 30B-A3B Instruct 2507 · 8-bit Alibaba52±1†local~41GB262kOllama › Qwen/Qwen3-Next-80B-A3B-Instruct model cardQwen/Qwen3-30B-A3B-Instruct-2507 model card
234Llama 3.3 Nemotron Super 49B v1.5 · 4-bit NVIDIA52±2†local~33GB131kllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
236Olmo 3.1 32B Think · 8-bit Ai252±3†local~43GB66kOllama › Olmo 3
237QwQ-32B · 8-bit Alibaba52±1local~43GB131kOllama › Aider polyglot benchmark runsCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WebApp1K models leaderboard (onekq-ai)
239Falcon-H1R 7B · 8-bit TII52±4*†local~12GB262kllama.cpp › Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling
241Olmo 3 32B Think · 8-bit Ai251±3†local~43GB66kOllama › Olmo 3
242Granite 4.1 30B · 4-bit IBM51±2†local~19GBOllama › ibm-granite/granite-4.1-8b model card
243s1.1 32B · 8-bit Simple Scaling51±9*local~38GB33kllama.cpp › open-thoughts/OpenThinker-32B model card
244Qwen3 Coder 30B-A3B Instruct · 8-bit Alibaba51±2local~41GB160kOllama › Qwen/Qwen3-Coder-30B-A3B-Instruct model cardKwaipilot/KAT-Coder-V2.5-Dev model cardRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
247Qwen3-VL 8B Thinking · 4-bit Alibaba51±2*†local~12GB256kLM Studio › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
250Qwen2.5 Coder 14B Instruct · 8-bit Alibaba51±1local~21GB131kOllama › Aider code editing benchmark runs (old edit leaderboard)Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
255Qwen2.5 72B Instruct · 4-bit Alibaba51±1local~48GB33kllama.cpp › Cohere — Command A technical reportAider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)Microsoft Research — Phi-4 Technical ReportWildEval ZeroEval published results+2
257Nemotron Nano 9B v2 · 8-bit NVIDIA50±1local~13GB131kllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning ModelsOlmo 3NVIDIA vendor-reported benchmark scores for its own Nemotron models (Hugging Face model cards)
260Mistral Large 2 (2407) · 2-bit Mistral50±1local~45GB128kllama.cpp › Aider code editing benchmark runs (old edit leaderboard)Vector Institute Evaluation Leaderboard (Inspect Evals runs)WildEval ZeroEval published resultsWebApp1K models leaderboard (onekq-ai)
261Command A (03-2025) · 2-bit Cohere50±2†local~37GB256kllama.cpp › Cohere — Command A technical report
265Qwen3 4B Thinking 2507 · 8-bit Alibaba50±6*†local~7.1GB262kOllama › Qwen/Qwen3-4B-Thinking-2507 model card
267Ministral 3 14B Reasoning 2512 · 4-bit Mistral50±4local~10GB262kLM Studio › JetBrains/Mellum2-12B-A2.5B-Thinking model cardmistralai/Ministral-3-8B-Reasoning-2512 model cardmistralai/Ministral-3-3B-Reasoning-2512 model card
268Phi-4 14B · 8-bit Microsoft50±1local~21GB16kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)Microsoft Research — Phi-4 Technical ReportMicrosoft — Phi-4-reasoning Technical Reportmicrosoft/Phi-4-reasoning model cardmicrosoft/phi-4 model card
269Granite 4.2 3B · 8-bit IBM50±2local~5.6GB131kllama.cpp › openbmb/MiniCPM5-2B model cardibm-granite/granite-4.2-30b model card
270Mellum2 12B-A2.5B Thinking · 8-bit JetBrains50±2†local~15GB131kllama.cpp › JetBrains/Mellum2.1-12B-A2.5B-Thinking model cardJetBrains/Mellum2-12B-A2.5B-Thinking model card
272EXAONE 4.0 32B · 8-bit LG AI Research49±3*†local~42GB131kllama.cpp › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)
273Olmo 3 7B Think · 8-bit Ai249±2local~11GB66kOllama › Olmo 3JetBrains/Mellum2-12B-A2.5B-Thinking model card
274Gemma 2 27B IT · 8-bit Google49±2local~36GB8kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WildEval ZeroEval published results
276InternLM2.5 20B Chat · 8-bit Shanghai AI Laboratory (InternLM)49±2local~27GB33kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
277Qwen3.5 4B · 4-bit Alibaba49±1local~6.9GB262kLM Studio › openbmb/MiniCPM5-2B model cardXHToken/Spark-X2.5-4B model cardJetBrains/Mellum2-12B-A2.5B-Instruct model cardLiquidAI/LFM2.5-2.6B model cardQwen/Qwen3.5-4B model cardJetBrains/Mellum2-12B-A2.5B-Thinking model card+1
278Llama 3.3 70B Instruct · 4-bit Meta49±1local~46GB128kOllama › Cohere — Command A technical reportAider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)Microsoft Research — Phi-4 Technical ReportWebApp1K models leaderboard (onekq-ai)
279Mistral Small 3 24B (2501) · 8-bit Mistral49±4*local~32GB32kOllama › WebApp1K models leaderboard (onekq-ai)
281DeepSeek R1 Distill Qwen 14B · 8-bit DeepSeek49±2local~21GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)deepseek-ai/DeepSeek-R1 model card
282Qwen2.5 14B Instruct · 8-bit Alibaba49±1local~21GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)Microsoft Research — Phi-4 Technical Reportmicrosoft/phi-4 model card
283Mistral Small Instruct 22B (2409) · 8-bit Mistral AI48±2local~30GB32kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
285Qwen3 30B-A3B (Apr 2025, hybrid) · 8-bit Alibaba48±1†local~41GB33kOllama › Qwen/Qwen3-4B-Instruct-2507 model cardQwen/Qwen3-30B-A3B-Instruct-2507 model cardQwen/Qwen3-4B-Thinking-2507 model cardQwen/Qwen3-30B-A3B-Thinking-2507 model card
286CodeQwen1.5 7B Chat · 8-bit Alibaba48±1local~11GB64kOllama › EvalPlus / EvalPerf published evaluation runsQwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
287OpenCoder-8B-Instruct · 16-bit INF (infly)48±1local~18GBhow to run › EvalPlus / EvalPerf published evaluation runs
288DeepSeek R1 Distill Llama 70B · 4-bit DeepSeek48±2local~46GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)deepseek-ai/DeepSeek-R1 model card
289Llama 3.1 70B Instruct · 4-bit Meta48±1local~46GB128kOllama › Aider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)Vector Institute Evaluation Leaderboard (Inspect Evals runs)WildEval ZeroEval published resultsWebApp1K models leaderboard (onekq-ai)+1
290Qwen3 8B · 8-bit Alibaba48±2local~12GB33kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling
292Qwen2.5 Coder 7B Instruct · 8-bit Alibaba48±1local~12GB131kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Aider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)
293DeepSeek R1 Distill Qwen 32B · 8-bit DeepSeek48±2local~43GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WebApp1K models leaderboard (onekq-ai)deepseek-ai/DeepSeek-R1 model cardopen-thoughts/OpenThinker-32B model card
295DeepSeek R1 0528 Qwen3 8B · 8-bit DeepSeek47±4local~12GB128kOllama › Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scalingdeepseek-ai/DeepSeek-R1-0528-Qwen3-8B model card
296Qwen3-VL 4B Thinking · 8-bit Alibaba47±2*†local~6.9GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
297Nemotron 3 Nano 30B-A3B · 8-bit NVIDIA47±3†local~42GB262kOllama › NVIDIA — Nemotron-Cascade 2 technical report
299DeepSeek LLM 67B Chat · 4-bit DeepSeek47±2local~44GB4kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
300OpenThinker3 7B · 8-bit Open Thoughts47±2local~10GB33kllama.cpp › Olmo 3
302Qwen2 72B Instruct · 4-bit Alibaba47±1local~48GB131kOllama › Aider code editing benchmark runs (old edit leaderboard)LiveCodeBench official website results (2024-12-26)WildEval ZeroEval published resultsSciCode leaderboard (repository README)
304Phi-3 Medium 128K Instruct · 8-bit Microsoft47±2local~20GB128kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
305OpenCodeInterpreter-DS-33B · 8-bit M-A-P47±1local~39GBllama.cpp › EvalPlus / EvalPerf published evaluation runs
307Gemma 4 E4B · 4-bit Google47±1local~11GB131kLM Studio › openbmb/MiniCPM5-2B model cardJetBrains/Mellum2.1-12B-A2.5B-Thinking model cardGemma Team — Gemma 4 Technical ReportLiquidAI/LFM2.5-2.6B model cardXHToken/Spark-X2.5-4B model card
308DeepSeek-Coder 33B Instruct · 8-bit DeepSeek46±1local~39GB16kllama.cpp › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)CRUXEval published evaluation results (authors' generations and scores)
309Codestral 22B v0.1 · 8-bit Mistral AI46±1local~30GB32kOllama › EvalPlus / EvalPerf published evaluation runsAider code editing benchmark runs (old edit leaderboard)
310WizardCoder 33B V1.1 · 8-bit WizardLM46±1local~39GB16kllama.cpp › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsWizardLM team vendor-reported scores for its own WizardCoder models (Hugging Face card)
311Gemma 2 9B IT · 8-bit Google46±2local~14GB8kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WildEval ZeroEval published results
313Qwen2.5 7B Instruct · 8-bit Alibaba46±1local~12GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)WildEval ZeroEval published results
314Mistral NeMo Instruct (2407) · 8-bit Mistral46±2local~18GB128kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
315Llama 3.1 Nemotron 70B Instruct · 4-bit NVIDIA46±2*local~47GB128kOllama › Aider code editing benchmark runs (old edit leaderboard)WebApp1K models leaderboard (onekq-ai)
316Llama 3 70B Instruct · 4-bit Meta46±1local~47GB8kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsLiveCodeBench official website results (2024-12-26)WildEval ZeroEval published resultsWebApp1K models leaderboard (onekq-ai)SciCode leaderboard (repository README)
317OpenCodeInterpreter-DS-6.7B · 16-bit M-A-P46±1local~16GBhow to run › EvalPlus / EvalPerf published evaluation runs
320speechless-codellama-34b-v2.0 · 8-bit uukuguy46±1local~39GBllama.cpp › EvalPlus / EvalPerf published evaluation runs
321Magicoder-S-DS 6.7B · 4-bit UIUC (ise-uiuc)46±1local~30GBhow to run › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
322Artigenz-Coder-DS-6.7B · 16-bit Artigenz46±1local~16GBhow to run › EvalPlus / EvalPerf published evaluation runs
325Yi 1.5 9B Chat · 8-bit 01.AI45±2local~13GB4kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WildEval ZeroEval published results
326Nemotron 3 Nano 4B · 8-bit NVIDIA45±2local~7GB262kOllama › openbmb/MiniCPM5-2B model cardNVIDIA vendor-reported benchmark scores for its own Nemotron models (Hugging Face model cards)
327Pixtral 12B (2409) · 8-bit Mistral45±3*local~15GBllama.cpp › Pixtral 12B (Mistral AI, arXiv 2410.07073v2)VisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)
329Granite 4.1 3B · 4-bit IBM45±2†local~3.7GBOllama › ibm-granite/granite-4.1-8b model card
330XwinCoder-34B · 8-bit Xwin-LM45±1local~39GBllama.cpp › EvalPlus / EvalPerf published evaluation runs
331StarChat2 15B v0.1 · 16-bit Hugging Face45±1local~35GB16khow to run › EvalPlus / EvalPerf published evaluation runs
332Code Millenials 34B · 8-bit Bud Ecosystem45±1local~39GBllama.cpp › EvalPlus / EvalPerf published evaluation runs
333WizardLM 70B V1.0 · 4-bit Microsoft45±2local~45GBOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
334DeepSeek-Coder 6.7B Instruct · 8-bit DeepSeek45±1local~9.1GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsLiveCodeBench official website results (2024-12-26)CRUXEval published evaluation results (authors' generations and scores)
335AceReason-Nemotron-1.0-14B · 8-bit NVIDIA45±2†local~18GBllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
336WaveCoder-Ultra-6.7B · 4-bit Microsoft45±1local~30GBhow to run › EvalPlus / EvalPerf published evaluation runs
338Yi-Coder 9B Chat · 8-bit 01.AI45±1local~13GB128kOllama › Aider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)
339MiniCPM3-4B · 4-bit OpenBMB45±2*local~4.1GB32kllama.cpp › LiveCodeBench official website results (2024-12-26)
340Gemma 3 27B · 8-bit Google44±1local~37GB131kOllama › Google DeepMind Gemma 3 270M IT model cardAider polyglot benchmark runsWebApp1K models leaderboard (onekq-ai)Gemma Team — Gemma 4 Technical ReportVisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
341Ministral 3 14B · 8-bit Mistral44±4*local~20GB262kOllama › JetBrains/Mellum2-12B-A2.5B-Instruct model card
342OpenThinker2 7B · 4-bit Open Thoughts44±9*†local~6.4GB33kllama.cpp › open-thoughts/OpenThinker2-7B model card
343DeepSeek-Coder 7B Instruct v1.5 · 16-bit DeepSeek44±1local~16GBhow to run › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
345Qwen3 4B · 8-bit Alibaba44±2local~7.1GB33kOllama › Qwen/Qwen3-4B-Instruct-2507 model cardQwen/Qwen3-4B-Thinking-2507 model cardHuggingFaceTB/SmolLM3-3B model card (Evaluation section)
346Gemma 3 12B · 8-bit Google44±2local~17GB131kOllama › Google DeepMind Gemma 3 270M IT model cardVisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)
347Yi 1.5 34B Chat · 8-bit 01.AI44±2local~45GB4kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WildEval ZeroEval published results
349Qwen3-VL 32B Instruct · 8-bit Alibaba44±2*†local~39GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
350OpenChat 3.5 0106 · 8-bit OpenChat44±1local~11GB8kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
351Athene-70B · 4-bit Nexusflow43±3*local~46GBllama.cpp › WildEval ZeroEval published results
352Granite 34B Code Instruct (8K) · 4-bit IBM43±1local~24GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
353Phind CodeLlama 34B v2 · 8-bit Phind43±1local~39GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
354WhiteRabbitNeo-33B-v1 · 8-bit WhiteRabbitNeo43±1local~39GBllama.cpp › EvalPlus / EvalPerf published evaluation runs
355Magicoder-S-CL-7B · 4-bit UIUC (ise-uiuc)43±1local~30GBhow to run › EvalPlus / EvalPerf published evaluation runs
356Qwen2.5 Coder 3B Instruct · 4-bit Alibaba43±1local~5GB33kLM Studio › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Aider code editing benchmark runs (old edit leaderboard)
357Code Llama 70B Instruct · 4-bit Meta43±1local~46GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runs
358Qwen3-VL 30B-A3B Instruct · 8-bit Alibaba43±2*†local~37GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
359speechless-coder-ds-6.7b · 4-bit uukuguy43±1local~16GBhow to run › EvalPlus / EvalPerf published evaluation runs
360Spark-X2.5 1.7B · 16-bit iFLYTEK (SparkLLM)43±3†local~5.1GB1MOllama › XHToken/Spark-X2.5-4B model card
362WizardCoder Python 34B V1.0 · 8-bit WizardLM43±1local~39GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsWizardLM team vendor-reported scores for its own WizardCoder models (Hugging Face card)CRUXEval published evaluation results (authors' generations and scores)
363Phi-4-mini 3.8B · 8-bit Microsoft43±2†local~6.9GB128kOllama › Microsoft — Phi-4-Mini Technical Report
365AceReason-Nemotron-1.1-7B · 8-bit NVIDIA42±2†local~10GBllama.cpp › NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models
367DeepSeek-Coder-V2 Lite Instruct (16B-A2.4B) · 8-bit DeepSeek42±2local~22GB128kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)WebApp1K models leaderboard (onekq-ai)
368Solar Open 100B (102B-A12B) · 2-bit Upstage42±3*†local~41GB131kllama.cpp › Solar Open 2 Technical Report (Upstage, arXiv:2607.20062v2)
369Qwen3 4B Instruct 2507 · 8-bit Alibaba42±2local~7.1GB262kOllama › Qwen/Qwen3-4B-Instruct-2507 model cardQwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
373Mellum2 12B-A2.5B Instruct · 8-bit JetBrains42±4*†local~15GB131kllama.cpp › JetBrains/Mellum2-12B-A2.5B-Instruct model card
374Llama 3.1 8B Instruct · 8-bit Meta42±1local~12GB128kOllama › EvalPlus / EvalPerf published evaluation runsAider code editing benchmark runs (old edit leaderboard)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)IBM Granite 3.0 Language Models technical report (GitHub)LiveCodeBench official website results (2024-12-26)WildEval ZeroEval published results+2
375Llama 3.2 11B Vision Instruct · 8-bit Meta42±3*local~16GB128kOllama › Pixtral 12B (Mistral AI, arXiv 2410.07073v2)
376speechless-starcoder2-15b · 4-bit uukuguy42±1local~35GBhow to run › EvalPlus / EvalPerf published evaluation runs
377Granite 3.0 8B Instruct · 8-bit IBM42±2local~12GB4kOllama › IBM Granite 3.0 Language Models technical report (GitHub)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
378Qwen2-VL 7B Instruct · 8-bit Alibaba42±3*local~11GBllama.cpp › Pixtral 12B (Mistral AI, arXiv 2410.07073v2)
379Phi-3-mini-128k-instruct (Apr 2024) · 16-bit Microsoft42±2local~9.5GBhow to run › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
380Qwen2.5 3B Instruct · 8-bit Alibaba42±2local~5.9GB33kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)WildEval ZeroEval published results
381Qwen1.5 72B Chat · 4-bit Alibaba41±1local~48GB32kllama.cpp › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
382Phi-3-mini-4k-instruct (June 2024 update) · 4-bit Microsoft41±3*local~5.8GBLM Studio › WildEval ZeroEval published results
383Qwen3-VL 8B Instruct · 4-bit Alibaba41±2*†local~12GB256kLM Studio › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
384StarCoder2-15B-Instruct-v0.1 · 16-bit BigCode41±1local~35GB16khow to run › EvalPlus / EvalPerf published evaluation runs
386Granite 8B Code Instruct (4K) · 4-bit IBM41±2†local~6.6GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
387Granite 20B Code Instruct (8K) · 4-bit IBM41±2†local~15GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
388Phi-3.5-mini-instruct · 8-bit Microsoft41±2local~6.8GB128kOllama › Microsoft — Phi-4-Mini Technical ReportWildEval ZeroEval published results
389Falcon3 7B Instruct · 8-bit TII41±2local~11GB32kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
390Qwen3-VL 4B Instruct · 8-bit Alibaba41±2*†local~6.9GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
391Qwen1.5 110B Chat · 2-bit Alibaba40±2*local~45GB32kllama.cpp › Aider code editing benchmark runs (old edit leaderboard)
392Code Llama 70B Python · 4-bit Meta40±1local~45GB16kllama.cpp › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runs
393Dolphin 2.6 Mixtral 8x7B · 4-bit Cognitive Computations40±1local~29GBllama.cpp › EvalPlus / EvalPerf published evaluation runs
394Command R (08-2024) · 8-bit Cohere40±1local~43GB128kOllama › Aider code editing benchmark runs (old edit leaderboard)Cohere — Command A technical report
395DeepSeek-Coder 1.3B Instruct · 8-bit DeepSeek40±1local~3GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)LiveCodeBench official website results (2024-12-26)CRUXEval published evaluation results (authors' generations and scores)
396Command R Plus (04-2024) · 2-bit Cohere40±1local~43GB128kllama.cpp › EvalPlus / EvalPerf published evaluation runsAider code editing benchmark runs (old edit leaderboard)Vector Institute Evaluation Leaderboard (Inspect Evals runs)
397GLM-4.7-Flash · 8-bit Z.ai40±6*local~35GB200kllama.cpp › Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode
398Olmo 3 7B Instruct · 8-bit Ai239±2local~11GB66kOllama › Olmo 3JetBrains/Mellum2-12B-A2.5B-Instruct model card
399OpenChat 3.5 · 8-bit OpenChat39±3*†local~11GB8kOllama › OpenChat 3.5 model card (openchat/openchat_3.5)
400Llama 3 8B Instruct · 8-bit Meta39±1local~12GB8kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsWildEval ZeroEval published resultsLiveCodeBench official website results (2024-12-26)WebApp1K models leaderboard (onekq-ai)
401Code Llama 70B (base) · 4-bit Meta39±1local~45GB16kllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runs
402Qwen2 7B Instruct · 8-bit Alibaba39±3*local~12GB131kOllama › WildEval ZeroEval published results
403Falcon3 10B Instruct · 8-bit TII39±2local~15GB32kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
404Code Tulu 2 13B · 16-bit Allen Institute for AI39±2*†local~29GBhow to run › Tulu 2 — base and instruction-tuned evaluations
405WizardCoder Python 13B V1.0 · 8-bit WizardLM39±3local~16GB16kOllama › WizardLM team vendor-reported scores for its own WizardCoder models (Hugging Face card)CRUXEval published evaluation results (authors' generations and scores)
406Qwen2.5 Coder 1.5B Instruct · 8-bit Alibaba39±1local~3.5GB33kllama.cpp › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Aider code editing benchmark runs (old edit leaderboard)
407Gemma 4 E2B · 4-bit Google39±2local~7.7GB131kLM Studio › openbmb/MiniCPM5-2B model cardGemma Team — Gemma 4 Technical ReportLiquidAI/LFM2.5-2.6B model cardXHToken/Spark-X2.5-4B model card
408CodeGemma 7B IT (v1.0) · 4-bit Google39±1local~12GBLM Studio › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCodeGemma: Open Code Models Based on Gemma — authors' own runs
409LFM2.5 2.6B · 4-bit Liquid AI38±2local~4.5GB131kLM Studio › openbmb/MiniCPM5-2B model cardLiquidAI/LFM2.5-2.6B model card
410Granite 34B Code Base (8K) · 4-bit IBM38±1†local~24GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
412speechless-starcoder2-7b · 4-bit uukuguy38±1local~17GBhow to run › EvalPlus / EvalPerf published evaluation runs
414Code Llama 34B Python · 8-bit Meta38±1local~39GB16kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
415WizardCoder-15B-V1.0 · 4-bit WizardLM38±1local~34GBhow to run › EvalPlus / EvalPerf published evaluation runs
416Qwen2.5 1.5B Instruct · 8-bit Alibaba37±2local~4GB33kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)SmolLM2 paper (Hugging Face) — authors' own instruct-model evaluations
417Hermes 3 Llama 3.1 8B · 8-bit Nous Research37±2*local~12GB128kOllama › Aider code editing benchmark runs (old edit leaderboard)
418speechless-coding-7b-16k-tora · 16-bit uukuguy37±1local~16GBhow to run › EvalPlus / EvalPerf published evaluation runs
419Qwen3 1.7B · 8-bit Alibaba37±2*local~4.6GB33kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)HuggingFaceTB/SmolLM3-3B model card (Evaluation section)
420OpenHermes 2.5 Mistral 7B · 16-bit Nous Research37±3*local~17GBLM Studio › OpenChat 3.5 model card (openchat/openchat_3.5)
421Gemma 3 4B · 8-bit Google37±2local~7.4GB131kOllama › Google DeepMind Gemma 3 270M IT model cardVisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)
422Code Llama 34B Instruct · 8-bit Meta37±1local~45GB16kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsTulu 2 — base and instruction-tuned evaluations
423Command R7B (12-2024) · 8-bit Cohere37±2†local~12GB128kOllama › Cohere — Command A technical report
424OpenHermes-2.5-Code-290k-13B · 16-bit ajibawa-202337±1local~29GBhow to run › EvalPlus / EvalPerf published evaluation runs
425WizardLM 13B V1.2 · 4-bit Microsoft36±3local~29GBhow to run › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
426Qwen1.5 14B Chat · 8-bit Alibaba36±1*local~20GB32kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
427Granite 3.0 2B Instruct · 8-bit IBM36±2local~5.3GB4kOllama › IBM Granite 3.0 Language Models technical report (GitHub)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
428DeepSeek-Coder 33B Base · 8-bit DeepSeek36±1local~39GB16kllama.cpp › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
429Qwen3-VL 2B Thinking · 8-bit Alibaba36±3*†local~4.2GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
430CodeQwen1.5 7B (base) · 16-bit Alibaba36±1local~17GB66khow to run › EvalPlus / EvalPerf published evaluation runs
431Code Llama 34B (base) · 8-bit Meta36±1local~39GB16kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsTulu 2 — base and instruction-tuned evaluationsCRUXEval published evaluation results (authors' generations and scores)
432Llama 3.2 3B Instruct · 8-bit Meta36±1local~6.1GB131kOllama › Aider code editing benchmark runs (old edit leaderboard)IBM Granite 3.0 Language Models technical report (GitHub)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
433Granite 3B Code Instruct (2K) · 4-bit IBM36±2†local~3.7GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
434Phi-2 (2.7B) · 4-bit Microsoft35±1local~7.3GB2khow to run › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsMiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)CRUXEval published evaluation results (authors' generations and scores)
435Falcon3 3B Instruct · 8-bit TII35±3local~6.1GB32kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
436SmolLM3 3B · 8-bit Hugging Face35±6*†local~5.9GB66kllama.cpp › HuggingFaceTB/SmolLM3-3B model card (Evaluation section)
437Tulu 2 DPO 70B · 4-bit Ai235±2*†local~45GBllama.cpp › Tulu 2 — base and instruction-tuned evaluations
438Qwen-14B-Chat · 16-bit Alibaba35±1*local~31GBhow to run › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
439Code Llama 13B Instruct · 8-bit Meta35±1local~16GB16kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsTulu 2 — base and instruction-tuned evaluations
440Code Tulu 2 7B · 16-bit Allen Institute for AI35±2*†local~16GBhow to run › Tulu 2 — base and instruction-tuned evaluations
441LFM2.5 8B-A1B · 8-bit Liquid AI35±2local~13GB128kOllama › openbmb/MiniCPM5-2B model card
442DeepSeek-V2 Lite Chat (16B-A2.4B) · 8-bit DeepSeek35±2local~22GB32kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)DeepSeek vendor-reported benchmark scores for its own models (Hugging Face model cards)LiveCodeBench official website results (2024-12-26)
443Tulu 2 70B · 4-bit Allen Institute for AI35±2*†local~45GBllama.cpp › Tulu 2 — base and instruction-tuned evaluations
444Granite 8B Code Base (4K) · 4-bit IBM35±1†local~6.6GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
445DeepSeek-Coder 6.7B Base · 8-bit DeepSeek35±1local~9.1GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
446Mixtral 8x7B Instruct v0.1 · 4-bit Mistral34±1local~31GB32kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsWildEval ZeroEval published resultsWebApp1K models leaderboard (onekq-ai)
447MistralHermes-CodePro-7B-v1 · 16-bit beowolx34±1local~17GBhow to run › EvalPlus / EvalPerf published evaluation runs
448speechless-code-mistral-7b-v1.0 · 4-bit uukuguy34±1local~17GBhow to run › EvalPlus / EvalPerf published evaluation runs
449Code Llama 13B Python · 8-bit Meta34±1local~16GB16kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
450StarCoder2 15B · 8-bit BigCode34±1local~19GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runs
451Mixtral 8x7B v0.1 (base) · 4-bit Mistral AI34±1local~31GB32kOllama › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsMixtral of Experts — base-model evaluationsCRUXEval published evaluation results (authors' generations and scores)
452Xwin-LM 70B v0.1 · 4-bit Xwin-LM34±2*local~45GBllama.cpp › Tulu 2 — base and instruction-tuned evaluations
453CodeGemma 7B (v1.0) · 8-bit Google34±1local~11GBOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCodeGemma: Open Code Models Based on Gemma — authors' own runs
454Qwen1.5 7B Chat · 8-bit Alibaba34±1*local~12GB32kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
455EXAONE 3.5 2.4B Instruct · 8-bit LG AI Research34±3local~5.1GB33kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
456Granite 20B Code Base (8K) · 4-bit IBM33±2†local~15GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
457Code Llama 7B Instruct · 8-bit Meta33±1local~9.1GB16kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsTulu 2 — base and instruction-tuned evaluations
458Code Llama 7B Python · 8-bit Meta33±1local~9.1GB16kOllama › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
459Code Llama 13B (base) · 8-bit Meta33±1local~16GB16kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsTulu 2 — base and instruction-tuned evaluationsCRUXEval published evaluation results (authors' generations and scores)
460SOLAR 10.7B Instruct v1.0 · 8-bit Upstage33±1local~16GB4kOllama › EvalPlus / EvalPerf published evaluation runs
461Qwen2.5 Coder 0.5B Instruct · 4-bit Alibaba33±1local~2GB33kLM Studio › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)Aider code editing benchmark runs (old edit leaderboard)
462Gemma 1.1 7B IT · 8-bit Google32±1local~13GB8kOllama › EvalPlus / EvalPerf published evaluation runs
463Gemma 2 2B IT · 8-bit Google32±2local~5.3GB8kOllama › IBM Granite 3.0 Language Models technical report (GitHub)WildEval ZeroEval published results
464DeepSeek LLM 7B Chat · 4-bit DeepSeek32±3*†local~5.7GBOllama › DeepSeek vendor-reported benchmark scores for its own models (Hugging Face model cards)
465SmolLM2 1.7B Instruct · 8-bit Hugging Face32±4*†local~4.2GB8kOllama › SmolLM2 paper (Hugging Face) — authors' own instruct-model evaluations
466Mistral 7B Instruct v0.2 · 8-bit Mistral32±1local~11GB32kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
467Llama 3.2 1B Instruct · 4-bit Meta31±2local~5.2GB131kOllama › IBM Granite 3.0 Language Models technical report (GitHub)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)SmolLM2 paper (Hugging Face) — authors' own instruct-model evaluations
468xDAN-L1-Chat-RL-v1 · 4-bit xDAN-AI31±1local~32GBhow to run › EvalPlus / EvalPerf published evaluation runs
469Granite 3.0 3B-A800M Instruct · 8-bit IBM31±2*†local~5.3GB4kllama.cpp › IBM Granite 3.0 Language Models technical report (GitHub)
470Code Llama 7B (base) · 8-bit Meta31±1local~9.1GB16kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCode Llama: Open Foundation Models for Code — authors' own runsTulu 2 — base and instruction-tuned evaluationsMistral 7B — authors' own runs+1
471Llama 2 70B (base) · 4-bit Meta30±1local~45GB4kllama.cpp › Tulu 2 — base and instruction-tuned evaluationsMixtral of Experts — base-model evaluations
472Granite 3B Code Base (2K) · 4-bit IBM30±2†local~3.7GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
473StarCoder2 7B · 8-bit BigCode30±1local~9.5GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
474Qwen3-VL 2B Instruct · 8-bit Alibaba30±3*†local~4.2GB256kOllama › Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)
475StarCoder 15.5B · 8-bit BigCode30±1local~19GBllama.cpp › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
476Gemma 7B (base) · 8-bit Google30±1local~11GB8kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCodeGemma: Open Code Models Based on Gemma — authors' own runsGemma report — original evaluations
477Llama 3 8B (base) · 16-bit Meta30±1local~18GBhow to run › EvalPlus / EvalPerf published evaluation runs
478StarCoder2 3B · 4-bit BigCode29±1local~5.1GB16kLM Studio › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
479StarCoderBase 15.5B · 8-bit BigCode29±1local~19GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
480Mistral 7B Instruct v0.1 · 8-bit Mistral29±1*local~9.6GB8kllama.cpp › Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs
481Llama 2 70B Chat · 4-bit Meta28±2*local~46GB4kOllama › Tulu 2 — base and instruction-tuned evaluations
483DeepSeek-Coder 1.3B Base · 8-bit DeepSeek28±1local~3GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
484Stable Code 3B · 8-bit Stability AI28±1local~4.7GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runs
485OpenThinker 7B · 8-bit Open Thoughts28±8*†local~10GB33kllama.cpp › open-thoughts/OpenThinker2-7B model cardopen-thoughts/OpenThinker-7B model card
486Mistral 7B v0.1 (base) · 16-bit Mistral AI28±1local~17GB8khow to run › EvalPlus / EvalPerf published evaluation runsMixtral of Experts — base-model evaluationsMiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)Mistral 7B — authors' own runsGemma report — original evaluations
487Gemma 7B IT · 8-bit Google28±1local~13GB8kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
488Zephyr 7B beta · 8-bit Hugging Face28±1local~11GBOllama › EvalPlus / EvalPerf published evaluation runsTulu 2 — base and instruction-tuned evaluationsOpenChat 3.5 model card (openchat/openchat_3.5)
489DeepSeek R1 Distill Qwen 7B · 8-bit DeepSeek27±3local~12GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)deepseek-ai/DeepSeek-R1 model cardopen-thoughts/OpenThinker2-7B model cardopen-thoughts/OpenThinker-7B model card
490DeepSeek R1 Distill Llama 8B · 8-bit DeepSeek27±4local~12GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)deepseek-ai/DeepSeek-R1 model card
491Tulu 2 13B · 16-bit Allen Institute for AI27±2*†local~29GBhow to run › Tulu 2 — base and instruction-tuned evaluations
492Tulu 2 DPO 13B · 16-bit Allen Institute for AI27±2*†local~29GBhow to run › Tulu 2 — base and instruction-tuned evaluations
493Granite 3.0 1B-A400M Instruct · 8-bit IBM27±2*†local~3GB4kllama.cpp › IBM Granite 3.0 Language Models technical report (GitHub)
494CodeGemma 2B (v1.0) · 8-bit Google27±1local~4.3GBOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCodeGemma: Open Code Models Based on Gemma — authors' own runs
496Falcon 40B (base) · 4-bit TII26±2*local~28GB2kllama.cpp › MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)
497StarCoderBase 7B · 4-bit BigCode26±1local~33GBhow to run › EvalPlus / EvalPerf published evaluation runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCRUXEval published evaluation results (authors' generations and scores)
498Falcon3 1B Instruct · 8-bit TII26±4local~4.1GB8kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
499Gemma 3 1B · 4-bit Google25±2†local~3GB32kLM Studio › Google DeepMind Gemma 3 270M IT model card
500Gemma 2B (base) · 8-bit Google25±1local~4.3GB8kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsCodeGemma: Open Code Models Based on Gemma — authors' own runsGemma report — original evaluations
501Qwen3.5 2B · 8-bit Alibaba24±3local~4.9GB262kOllama › openbmb/MiniCPM5-2B model cardXHToken/Spark-X2.5-4B model card
502Gemma 1.1 2B IT · 8-bit Google24±1local~5.2GB8kOllama › EvalPlus / EvalPerf published evaluation runs
503Tulu 2 DPO 7B · 16-bit Allen Institute for AI23±2*†local~16GBhow to run › Tulu 2 — base and instruction-tuned evaluations
504StarCoderBase 3B · 4-bit BigCode23±1local~14GBhow to run › EvalPlus / EvalPerf published evaluation runsStarCoder 2 and The Stack v2: The Next Generation — authors' own runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
506Llama 2 13B (base) · 8-bit Meta22±1local~16GB4kllama.cpp › Tulu 2 — base and instruction-tuned evaluationsMixtral of Experts — base-model evaluationsMiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)Mistral 7B — authors' own runs
507Tulu 2 7B · 16-bit Allen Institute for AI22±2*†local~16GBhow to run › Tulu 2 — base and instruction-tuned evaluations
508Gemma 2B IT · 8-bit Google22±1local~5.2GB8kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs
510Llama 2 13B Chat · 8-bit Meta21±2*local~18GB4kOllama › Tulu 2 — base and instruction-tuned evaluations
512OpenAssistant SFT-4 Pythia 12B · 4-bit LAION (OpenAssistant)20±2*local~27GBhow to run › InstructEval (DeCLaRe Lab) README results table
513Phi-1 (1.3B) · 4-bit Microsoft19±5*local~4.5GB2khow to run › CRUXEval published evaluation results (authors' generations and scores)
514GPT-J 6B · 4-bit EleutherAI19±1*local~14GB2khow to run › EvalPlus / EvalPerf published evaluation runs
515Llama 2 7B (base) · 8-bit Meta19±1local~9.1GB4kllama.cpp › Tulu 2 — base and instruction-tuned evaluationsMixtral of Experts — base-model evaluationsMiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)Mistral 7B — authors' own runs
517Qwen3 0.6B · 8-bit Alibaba17±3*local~3GB33kOllama › Olmo 3
518Llama 2 7B Chat · 8-bit Meta16±2*local~11GB4kOllama › Tulu 2 — base and instruction-tuned evaluations
520GPT-Neo 2.7B · 4-bit EleutherAI14±1*local~13GB2khow to run › EvalPlus / EvalPerf published evaluation runs
521DeepSeek R1 Distill Qwen 1.5B · 8-bit DeepSeek14±5local~4.3GB131kOllama › CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)deepseek-ai/DeepSeek-R1 model card
523StableLM-Base-Alpha 7B · 4-bit Stability AI5±2local~35GB4khow to run › EvalPlus / EvalPerf published evaluation runsInstructEval (DeCLaRe Lab) README results table
524Zyte-1B · 4-bit venkycs (aihub-app)3±2local~3.8GBhow to run › EvalPlus / EvalPerf published evaluation runs
—BLOOM 1.1B · fp16 BigSciencenot yet calibratedlocal~3.7GB2khow to run › BLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runs
—BLOOM 1.7B · fp16 BigSciencenot yet calibratedlocal~5.1GB2khow to run › BLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runs
—BLOOM 3B · fp16 BigSciencenot yet calibratedlocal~7.8GB2khow to run › BLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runs
—BLOOM 560M · fp16 BigSciencenot yet calibratedlocal~2.7GB2khow to run › BLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runs
—BLOOM 7.1B · fp16 BigSciencenot yet calibratedlocal~16GB2khow to run › BLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runs
—EXAONE Deep 2.4B · q4_k_m LG AI Researchnot yet calibratedlocal~3.2GB33kllama.cpp › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)
—EXAONE Deep 32B · q4_k_m LG AI Researchnot yet calibratedlocal~22GB33kllama.cpp › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)
—EXAONE Deep 7.8B · mlx-8bit LG AI Researchnot yet calibratedlocal~10GB33kLM Studio › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)
—Gemma 3n E2B IT · q8_0 Googlenot yet calibratedlocal~8.9GB32kOllama › Google DeepMind Gemma 3n E2B IT model card
—Gemma 3n E4B · q8_0 Googlenot yet calibratedlocal~12GB32kOllama › Google DeepMind Gemma 3n E2B IT model card
—Granite 4.0 1B · bf16 IBMnot yet calibratedlocal~6.1GB128kOllama › ibm-granite/granite-4.0-h-1b model card
—Granite 4.0 350M · bf16 IBMnot yet calibratedlocal~2.9GB32kOllama › ibm-granite/granite-4.0-h-1b model card
—Granite 4.0 H 1B · q8_0 IBMnot yet calibratedlocal~3.2GB131kOllama › ibm-granite/granite-4.0-h-1b model card
—Granite 4.0 H 350M · mlx-8bit IBMnot yet calibratedlocal~1.9GB33kLM Studio › ibm-granite/granite-4.0-h-1b model card
—Granite 4.0 H Micro 3B · mlx-8bit IBMnot yet calibratedlocal~5.1GB131kLM Studio › ibm-granite/granite-4.0-h-small model card
—Granite 4.0 H Small (32B-A9B) · Q8_0 IBMnot yet calibratedlocal~38GB131kllama.cpp › ibm-granite/granite-4.0-h-small model card
—Granite 4.0 H Tiny (7B-A1B) · mlx-8bit IBMnot yet calibratedlocal~9.3GB131kLM Studio › ibm-granite/granite-4.0-h-small model card
—Granite 4.0 Micro 3B · Q8_0 IBMnot yet calibratedlocal~5.3GB131kllama.cpp › ibm-granite/granite-4.0-h-small model card
—Jamba 1.5 Mini · Q4_K_M AI21 Labsnot yet calibratedlocal~34GBllama.cpp › Jamba-1.5: Hybrid Transformer-Mamba Models at Scale (AI21, arXiv 2408.12570v1)
—K2 Horizon 7B · gguf-q8_0 Institute of Foundation Models (MBZUAI)not scored yetlocal~13GB524kllama.cpp ›
—Llama 3.3 Nemotron Super 49B v1 · q4_k_m NVIDIAnot scored yetlocal~33GBllama.cpp ›
—Magistral Small 1.0 (2506) · q8_0 Mistralnot yet calibratedlocal~32GB128kOllama › mistralai/Magistral-Small-2509 model card
—Magistral Small 1.1 (2507) · Q8_0 Mistralnot yet calibratedlocal~28GB40kllama.cpp › mistralai/Magistral-Small-2509 model card
—Magistral Small 1.2 · gguf-q8_0 Mistralnot yet calibratedlocal~32GB131kllama.cpp › mistralai/Mistral-Small-4-119B-2603 model cardmistralai/Magistral-Small-2509 model card
—Ministral 8B (2410) · Q8_0 Mistralnot yet calibratedlocal~10GB128kllama.cpp › Mistral AI vendor-reported benchmark scores for its own models (Hugging Face model cards)
—Mistral 7B Instruct v0.3 · q8_0 Mistralnot yet calibratedlocal~11GB32kOllama › Mistral AI vendor-reported benchmark scores for its own models (Hugging Face model cards)
—Mistral Small 3.1 24B (2503) · q8_0 Mistralnot scored yetlocal~32GB128kOllama ›
—Mistral Small 3.2 24B · q8_0 Mistralnot scored yetlocal~32GB128kOllama ›
—Mistral Small 4 (119B-A6.5B) · ud-q2_k_xl Mistralnot yet calibratedlocal~40GB262kllama.cpp › mistralai/Mistral-Small-4-119B-2603 model card
—OLMo 7B · fp32 Allen Institute for AInot yet calibratedlocal~30GB2khow to run › OLMo: Accelerating the Science of Language Models — authors' own runs
—OLMo 7B Instruct · bf16 Ai2not yet calibratedlocal~16GBhow to run › OLMo: Accelerating the Science of Language Models — authors' own runs
—Qwen2 0.5B Instruct · q8_0 Alibabanot yet calibratedlocal~2.6GB32kOllama › Qwen2-0.5B-Instruct model card
—Qwen2 1.5B Instruct · q8_0 Alibabanot yet calibratedlocal~4GB32kOllama › Qwen2-0.5B-Instruct model card

+ 250 more don't fit your machine or budget · show them

+ 9 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them

572 of 598 catalogue models listed here · sources and licences

* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory