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

Best AI models to run locally on a 8GB Mac

A leaderboard of the best AI models you can run locally on a Mac with 8GB of memory. Spark-X2.5 4B is #1 right now, followed by MiniCPM5 2B and Qwen3.5 4B. Models are ranked by one score that combines 472 published benchmarks. Each model shows the download that fits and the command to run it.

136 models · scores combine 472 benchmarks · updated how it works · find a leaderboard · benchmark your machine
task[all tasks][agent tasks][coding][chat & writing][math][vision & documents][web search]
Models for 8GB RAM · Apple Silicon → ~5GB fast · local only · reset
+ more
chip[M1][M2][M3][M4]

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=8&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][$50][any]max per M

detecting your OS…

Score vs memory to run

#modelscore↓where↓speed↓RAM↓context↓get itsources
#modelscore↓where↓speed↓RAM↓context↓get itsources
153Spark-X2.5 4B · 4-bit iFLYTEK (SparkLLM)66±1†local~4.2GB1Mllama.cpp › XHToken/Spark-X2.5-4B model card
186MiniCPM5 2B · 4-bit OpenBMB64±1†local~4.3GB131kLM Studio › openbmb/MiniCPM5-2B model card
226Qwen3.5 4B · 4-bit Alibaba61±1local~5GB262kOllama › Qwen/Qwen3.5-9B model cardopenbmb/MiniCPM5-2B model cardXHToken/Spark-X2.5-4B model cardJetBrains/Mellum2-12B-A2.5B-Instruct model cardQwen/Qwen3.5-4B model cardLiquidAI/LFM2.5-2.6B model card+10
276Granite 4.2 3B · 4-bit IBM57±1local~4.3GB131kOllama › openbmb/MiniCPM5-2B model cardibm-granite/granite-4.2-30b model cardLMArena leaderboard dataset
283QED-Nano · 4-bit LM Provers57±2local~4.1GB262kllama.cpp › MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets
287Qwen3 4B Thinking 2507 · 4-bit Alibaba57±1local~4.6GB262kOllama › Qwen/Qwen3-4B-Thinking-2507 model cardMathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasetsMedmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 modelsBRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)Qwen/Qwen3.5-0.8B model cardKnowledge Index of Noah's Ark (arXiv:2606.05104v2)+2
295Qwen3-VL 4B Thinking · 4-bit Alibaba56±1local~5GB256kOllama › 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)ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)BEAR-Bench: A Bilingual Enterprise and Academic Reasoning Benchmark for Multimodal Models (arXiv:2608.17895v1)
317Qwen3 4B Instruct 2507 · 4-bit Alibaba55±1local~4.6GB262kOllama › Qwen/Qwen3-4B-Instruct-2507 model cardMarco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient UpcyclingQwen (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)Epoch AI — internal evaluationsRoblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau LeetcodeSynthesize and Reward: Reinforcement Learning for Multi-Step Tool Use in Live Environments (arXiv:2606.03892v2)+4
320Spark-X2.5 1.7B · 16-bit iFLYTEK (SparkLLM)55±1†local~5.1GB1MOllama › XHToken/Spark-X2.5-4B model card
335Nemotron 3 Nano 4B · 4-bit NVIDIA54±1local~4.5GB262kOllama › openbmb/MiniCPM5-2B model cardNVIDIA vendor-reported benchmark scores for its own Nemotron models (Hugging Face model cards)
368Qwen3-VL 4B Instruct · 4-bit Alibaba53±1local~5GB256kOllama › 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)ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)
370LFM2.5 2.6B · 4-bit Liquid AI53±1local~4.5GB131kLM Studio › openbmb/MiniCPM5-2B model cardLiquidAI/LFM2.5-2.6B model card
394Ministral 3 3B Reasoning 2512 · 4-bit Mistral51±1local~5.3GB262kLM Studio › mistralai/Ministral-3-8B-Reasoning-2512 model cardmistralai/Ministral-3-3B-Reasoning-2512 model cardMedmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 modelsEuroEval leaderboards (results collection), English generative leaderboard
395Qwen3 4B · 4-bit Alibaba51±1local~4.6GB33kOllama › Qwen/Qwen3-4B-Thinking-2507 model cardQwen/Qwen3-4B-Instruct-2507 model cardLFM2 Technical Report — Liquid AI own SmolLM3 evaluationHuggingFaceTB/SmolLM3-3B model card (Evaluation section)Epoch AI — internal evaluationsArena-Hard-Auto v2.0-Preview leaderboard (repository README)+8
409EXAONE Deep 2.4B · 4-bit LG AI Research50±2†local~3.2GB33kllama.cpp › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)
430Granite 4.1 3B · 4-bit IBM49±1†local~3.7GBOllama › ibm-granite/granite-4.1-8b model card
433LFM2.5 1.2B Thinking · 8-bit Liquid AI49±2local~3.5GB33kOllama › LiquidAI/LFM2.5-1.2B-Thinking model cardEuroEval leaderboards (results collection), English generative leaderboard
434Qwen3-VL 2B Thinking · 8-bit Alibaba49±1local~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)ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)
455T5 Large · 4-bit Google48±9local~4.6GB1khow to run › COPEN — EMNLP 2022 original zero-shot evaluations
465Qwen2.5 Coder 3B Instruct · 4-bit Alibaba48±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)
467MiniCPM3-4B · 4-bit OpenBMB48±2local~4.1GB32kllama.cpp › LiveCodeBench official website results (2024-12-26)Berkeley Function Calling Leaderboard result archive
468Bielik 4.5B v3.0 Instruct · 4-bit SpeakLeash48±2local~4.5GBllama.cpp › SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)Reassessing High-Performing LLMs on Polish Medical Exams: True Competence or Bias-Driven Performance? (arXiv:2606.12250v1)SpeakLeash Polish EQ-Bench leaderboardEuroEval leaderboards (results collection), English generative leaderboard
469Qwen3.5 2B · 8-bit Alibaba47±1local~4.9GB262kOllama › Qwen/Qwen3.5-2B model cardopenbmb/MiniCPM5-2B model cardXHToken/Spark-X2.5-4B model cardCohereLabs/North-Micro-Vision-Instruct model card (benchmark table)EuroEval leaderboards (results collection), English generative leaderboardMathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets+7
474Phi-4-mini-reasoning 3.8B · 4-bit Microsoft47±2local~4.8GB128kOllama › microsoft/Phi-4-mini-reasoning model cardBRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)EuroEval leaderboards (results collection), English generative leaderboard
478EXAONE 4.0 1.2B · 8-bit LG AI Research47±2†local~3.6GB66kllama.cpp › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)
481Gemma 4 E2B · 4-bit Google47±1local~5.2GB131kOllama › Gemma Team — Gemma 4 Technical Reportopenbmb/MiniCPM5-2B model cardXHToken/Spark-X2.5-4B model cardCohereLabs/North-Micro-Vision-Instruct model card (benchmark table)LiquidAI/LFM2.5-2.6B model cardEuroEval leaderboards (results collection), English generative leaderboard+6
488Granite 4.0 H Micro 3B · 4-bit IBM47±1local~5.1GB131kLM Studio › ibm-granite/granite-4.0-h-small model cardEuroEval leaderboards (results collection), English generative leaderboardVectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
492MedGemma 4B IT · 4-bit Google46±3local~5GBOllama › Reassessing High-Performing LLMs on Polish Medical Exams: True Competence or Bias-Driven Performance? (arXiv:2606.12250v1)BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)Open Portuguese LLM Leaderboard (archived)SurgiQ: A Large-Scale Multi-Domain Benchmark for Evaluating Surgical Understanding in Large Language Models (arXiv:2606.08071v1)
496Cogito v1 Preview Llama 3B · 4-bit Deep Cogito46±4local~5.1GB128kLM Studio › EuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
500Granite 4.0 Micro 3B · 8-bit IBM46±1local~5.3GB131kllama.cpp › ibm-granite/granite-4.0-h-small model cardEpoch AI — internal evaluationsEuroEval leaderboards (results collection), English generative leaderboardVectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
504T5 Base · 4-bit Google46±9local~2.4GB1khow to run › COPEN — EMNLP 2022 original zero-shot evaluations
507Qwen2.5-Math 1.5B Instruct · 4-bit Alibaba46±2†local~2.5GB4kllama.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)
524Ministral 3 3B · 4-bit Mistral45±1local~5.3GB262kLM Studio › Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient UpcyclingCohereLabs/North-Micro-Vision-Instruct model card (benchmark table)Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 modelsmistralai/Ministral-3-14B-Instruct-2512 model cardmistralai/Ministral-3-8B-Instruct-2512 model cardEuroEval leaderboards (results collection), English generative leaderboard+2
534Qwen2-Math 1.5B Instruct · 8-bit Alibaba45±2†local~3.2GB4kllama.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)
540Phi-3-mini-4k-instruct (June 2024 update) · 4-bit Microsoft44±1local~4GBllama.cpp › WildEval ZeroEval published resultsLMArena leaderboard datasetEuroEval leaderboards (results collection), English generative leaderboard
544Gemma 3 4B · 4-bit Google44±1local~4.7GB131kOllama › Google DeepMind Gemma 3 270M IT model cardLFM2 Technical Report — Liquid AI own SmolLM3 evaluationMedmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 modelsGemma Team — Gemma 3 Technical ReportMIMIC Clinical Decision Making leaderboardLMArena leaderboard dataset+11
547Qwen2.5-VL 3B Instruct · 8-bit Alibaba44±1local~5GBllama.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)ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)
553Qwen3-VL 2B Instruct · 8-bit Alibaba44±1local~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)CohereLabs/North-Micro-Vision-Instruct model card (benchmark table)ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)BEAR-Bench: A Bilingual Enterprise and Academic Reasoning Benchmark for Multimodal Models (arXiv:2608.17895v1)How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)
558Qwen3 1.7B · 8-bit Alibaba44±1local~4.6GB33kOllama › Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient UpcyclingQwen/Qwen3.5-2B 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)LiquidAI/LFM2-1.2B model cardHuggingFaceTB/SmolLM3-3B model card (Evaluation section)LiquidAI/LFM2.5-1.2B-Thinking model card+10
559Phi-4-mini 3.8B · 4-bit Microsoft44±1local~4.4GB128kOllama › microsoft/Phi-4-mini-instruct model cardMicrosoft — Phi-4-Mini Technical ReportDaekeun Kim Korean LLM benchmark runsSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)SpeakLeash Polish EQ-Bench leaderboardKorean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities+4
564LFM2.5 1.2B Instruct · 4-bit Liquid AI44±2local~2.8GB33kLM Studio › LiquidAI/LFM2.5-1.2B-Thinking model cardEuroEval leaderboards (results collection), English generative leaderboard
566Phi-3.5-mini-instruct · 4-bit Microsoft43±1local~4.4GB128kOllama › microsoft/Phi-4-mini-instruct model cardWildEval ZeroEval published resultsLiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasetsDaekeun Kim Korean LLM benchmark runsMicrosoft — Phi-4-Mini Technical ReportSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)+4
571Gemma 3n E2B IT · 4-bit Google43±2local~4.7GB32kllama.cpp › Google DeepMind Gemma 3n E2B IT model cardOpen Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)EuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
573T5 Small · 4-bit Google43±9local~1.8GB1khow to run › COPEN — EMNLP 2022 original zero-shot evaluations
575Qwen2.5 Coder 1.5B Instruct · 8-bit Alibaba43±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)
576LFM2 2.6B · 4-bit Liquid AI43±2local~4.3GB33kLM Studio › LFM2 Technical Report — Liquid AI own SmolLM3 evaluationEuroEval leaderboards (results collection), English generative leaderboard
577SmolLM3 3B · 4-bit Hugging Face43±1local~5GB66kLM Studio › LFM2 Technical Report — Liquid AI own SmolLM3 evaluationHuggingFaceTB/SmolLM3-3B model card (Evaluation section)Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 modelsInception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)EuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
580Qwen2.5 3B Instruct · 8-bit Alibaba43±1local~5.3GB33kllama.cpp › microsoft/Phi-4-mini-instruct model cardWildEval ZeroEval published resultsSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)SpeakLeash Polish EQ-Bench leaderboard+5
581DeepSeek-Coder 1.3B Instruct · 8-bit DeepSeek43±1local~3GB16kOllama › EvalPlus / EvalPerf published evaluation runsTop Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runsBigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4LiveCodeBench official website results (2024-12-26)CRUXEval published evaluation results (authors' generations and scores)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
613North Micro Vision Instruct (2.4B) · 4-bit Cohere41±2†local~3.8GB128khow to run › CohereLabs/North-Micro-Vision-Instruct model card (benchmark table)
615Granite 4.0 1B · 8-bit IBM41±1local~3.3GB128kllama.cpp › ibm-granite/granite-4.0-h-1b model cardLiquidAI/LFM2.5-1.2B-Thinking model cardEpoch AI — internal evaluations
619Granite 4.0 H 1B · 8-bit IBM40±1local~3.2GB131kOllama › ibm-granite/granite-4.0-h-1b model cardLiquidAI/LFM2.5-1.2B-Thinking model card
627LFM2 1.2B · 4-bit Liquid AI40±2local~2.8GB33kLM Studio › LiquidAI/LFM2-1.2B model cardEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
630Bielik 1.5B v3.0 Instruct · 16-bit SpeakLeash40±3local~4.8GBhow to run › SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)EuroEval leaderboards (results collection), English generative leaderboard
632Granite 3.3 2B Instruct · 8-bit IBM39±1†local~4.3GB128kllama.cpp › ibm-granite/granite-3.3-8b-instruct model card
641Falcon-H1 3B Instruct · 8-bit TII39±4local~5.1GB131kllama.cpp › Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)Open Portuguese LLM Leaderboard (archived)
642Phi-2 (2.7B) · 8-bit Microsoft39±1local~4.7GB2kllama.cpp › 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.)AlpacaEval 1.0 and 2.0 leaderboards (tatsu-lab/alpaca_eval repository)CRUXEval published evaluation results (authors' generations and scores)Open Portuguese LLM Leaderboard (archived)+2
643Granite 3.0 2B Instruct · 8-bit IBM39±1local~5.3GB4kOllama › IBM Granite 3.0 Language Models technical report (GitHub)LMArena leaderboard datasetCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)EuroEval leaderboards (results collection), English generative leaderboardVectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
648Llama 3.2 3B Instruct · 8-bit Meta38±1local~5.1GB131kllama.cpp › microsoft/Phi-4-mini-instruct model cardIBM Granite 3.0 Language Models technical report (GitHub)LFM2 Technical Report — Liquid AI own SmolLM3 evaluationAider code editing benchmark runs (old edit leaderboard)microsoft/Phi-4-mini-reasoning model cardMicrosoft — Phi-4-Mini Technical Report+11
654Phi-3-mini-4k-instruct (Apr 2024) · 4-bit Microsoft38±2local~4GBllama.cpp › MMLU-Pro author-released evaluation runsPolyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)LMArena leaderboard datasetSpeakLeash Polish EQ-Bench leaderboardOpen Portuguese LLM Leaderboard (archived)
657Granite 3B Code Instruct (2K) · 4-bit IBM38±2local~3.7GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsBigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4EuroEval leaderboards (results collection), English generative leaderboard
660Qwen3.5 0.8B · 8-bit Alibaba38±1local~3.1GB262kOllama › Qwen/Qwen3.5-2B model cardEuroEval leaderboards (results collection), English generative leaderboardQwen/Qwen3.5-0.8B model cardChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)Knowledge Index of Noah's Ark (arXiv:2606.05104v2)BEAR-Bench: A Bilingual Enterprise and Academic Reasoning Benchmark for Multimodal Models (arXiv:2608.17895v1)+2
665Yi-1.5 6B (base) · 4-bit 01.AI37±3local~5.1GB4kLM Studio › BigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4Open Portuguese LLM Leaderboard (archived)EuroEval leaderboards (results collection), English generative leaderboard
666Qwen2-VL 2B Instruct · 4-bit Alibaba37±1local~3.9GB33kLM 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)Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
667DeepSeek R1 Distill Qwen 1.5B · 8-bit DeepSeek37±2local~4.3GB131kOllama › deepseek-ai/DeepSeek-R1 model cardCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)Epoch AI — internal evaluationsTHINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models (arXiv:2505.22113v1)EuroEval leaderboards (results collection), English generative leaderboard+1
668Gemma 2 2B IT · 8-bit Google37±1local~5.3GB8kOllama › IBM Granite 3.0 Language Models technical report (GitHub)WildEval ZeroEval published resultsWildBench v2 leaderboard data (Ai2)LMArena leaderboard datasetSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)+4
669Qwen2.5 Coder 0.5B Instruct · 4-bit Alibaba37±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)
676Yi 1.5 6B Chat · 4-bit 01.AI37±2local~5.1GB4kLM Studio › BigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4WildBench v2 leaderboard data (Ai2)SpeakLeash Polish EQ-Bench leaderboardOpen Portuguese LLM Leaderboard (archived)Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
677Falcon3 3B Instruct · 8-bit TII37±3local~5.1GB32kllama.cpp › Berkeley Function Calling Leaderboard result archiveCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)Open Portuguese LLM Leaderboard (archived)
680Granite 3.1 2B Instruct · 8-bit IBM36±1local~5.2GB128kOllama › ibm-granite/granite-3.3-8b-instruct model cardibm-granite/granite-3.1-3b-a800m-instruct model cardLMArena leaderboard datasetSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
681EXAONE 3.5 2.4B Instruct · 8-bit LG AI Research36±2local~5.1GB33kOllama › LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)
684LFM2 700M · 4-bit Liquid AI36±2local~2.3GB33kLM Studio › LiquidAI/LFM2-1.2B model cardEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
685Granite 3.2 2B Instruct · 8-bit IBM36±1local~5.2GB131kOllama › ibm-granite/granite-3.3-8b-instruct model cardVectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
692Qwen2.5 1.5B Instruct · 8-bit Alibaba36±1local~4GB33kOllama › 2 OLMo 2 Furious (Team OLMo, Ai2) — authors' own evaluations of OLMo 2 Instruct and peer instruct modelsSmolLM2 paper (Hugging Face) — authors' own instruct-model evaluationsSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)SpeakLeash Polish EQ-Bench leaderboardOpen Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)+6
699Granite 3.0 3B-A800M Instruct · 8-bit IBM35±1local~5.3GB4kllama.cpp › IBM Granite 3.0 Language Models technical report (GitHub)EuroEval leaderboards (results collection), English generative leaderboard
702Granite 3.1 3B-A800M Instruct · 8-bit IBM35±2†local~5.2GB131kllama.cpp › ibm-granite/granite-3.1-3b-a800m-instruct model card
709Qwen3 0.6B · 8-bit Alibaba33±1local~3GB33kOllama › Olmo 3LiquidAI/LFM2-1.2B model cardBerkeley Function Calling Leaderboard result archiveOpen Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)EuroEval leaderboards (results collection), English generative leaderboard+4
712StarCoder2 3B · 4-bit BigCode33±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 runsBigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4
713Granite 3B Code Base (2K) · 4-bit IBM33±1local~3.7GBllama.cpp › Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsBigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4EuroEval leaderboards (results collection), English generative leaderboard
718Stable Code 3B · 8-bit Stability AI33±1local~4.7GB16kOllama › 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
719OLMo 2 1B Instruct (0425) · 8-bit Allen Institute for AI33±2local~3.2GB4kllama.cpp › 2 OLMo 2 Furious (Team OLMo, Ai2) — authors' own evaluations of OLMo 2 Instruct and peer instruct modelsOpen Portuguese LLM Leaderboard (archived)
720DeepSeek-Coder 1.3B Base · 8-bit DeepSeek32±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 runsBigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4CRUXEval published evaluation results (authors' generations and scores)
724Qwen1.5 4B Chat · 4-bit Alibaba32±2local~4.5GB32kOllama › LiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasetsPolyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)LMArena leaderboard datasetOpen Portuguese LLM Leaderboard (archived)EuroEval leaderboards (results collection), English generative leaderboard
726CodeGemma 2B (v1.0) · 8-bit Google32±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 runsBigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4
728Phi-1.5 (1.3B) · 4-bit Microsoft32±1local~4.5GB2khow to run › Stability AI — original StableLM zero-shot evaluation outputsTextbooks Are All You Need II: phi-1.5 technical report — authors' own runsOpen Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)Open Portuguese LLM Leaderboard (archived)
729Falcon-H1 1.5B Instruct · 16-bit TII31±4local~4.8GB131khow to run › Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)Open Portuguese LLM Leaderboard (archived)
731Qwen2 1.5B Instruct · 8-bit Alibaba31±1local~4GB32kOllama › WildEval ZeroEval published resultsQwen (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)Qwen2-0.5B-Instruct model cardLiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasetsSpeakLeash Polish EQ-Bench leaderboardOpen Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)+1
733Gemma 3 1B · 4-bit Google31±1local~3GB32kLM Studio › Google DeepMind Gemma 3 270M IT model card2 OLMo 2 Furious (Team OLMo, Ai2) — authors' own evaluations of OLMo 2 Instruct and peer instruct modelsLiquidAI/LFM2-1.2B model cardLiquidAI/LFM2.5-1.2B-Thinking model cardMIMIC Clinical Decision Making leaderboardGemma Team — Gemma 3 Technical Report+7
737LFM2 350M · 4-bit Liquid AI30±2local~1.9GB33kLM Studio › LiquidAI/LFM2-1.2B model cardEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
738SmolLM2 1.7B Instruct · 8-bit Hugging Face30±1local~4.2GB8kOllama › 2 OLMo 2 Furious (Team OLMo, Ai2) — authors' own evaluations of OLMo 2 Instruct and peer instruct modelsSmolLM2 paper (Hugging Face) — authors' own instruct-model evaluationsLMArena leaderboard datasetSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)EuroEval leaderboards (results collection), English generative leaderboard
739Llama 3.2 1B Instruct · 4-bit Meta30±1local~5.2GB131kOllama › IBM Granite 3.0 Language Models technical report (GitHub)2 OLMo 2 Furious (Team OLMo, Ai2) — authors' own evaluations of OLMo 2 Instruct and peer instruct modelsLiquidAI/LFM2-1.2B model cardSmolLM2 paper (Hugging Face) — authors' own instruct-model evaluationsLiquidAI/LFM2.5-1.2B-Thinking model cardLMArena leaderboard dataset+13
746StableLM-3B-4E1T · 8-bit Stability AI29±1local~4.7GB4kllama.cpp › Stability AI — original StableLM zero-shot evaluation outputsOpen Portuguese LLM Leaderboard (archived)
747Gemma 2B (base) · 8-bit Google29±1local~4.3GB8kOllama › EvalPlus / EvalPerf published evaluation runsGemma report — original evaluationsTop 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 runsHallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)+3
749Gemma 1.1 2B IT · 8-bit Google29±1local~5.2GB8kOllama › EvalPlus / EvalPerf published evaluation runsPolyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)LMArena leaderboard datasetSpeakLeash Polish EQ-Bench leaderboardArena-Hard-Auto v0.1 leaderboard (repository README, pre-v2)Open Portuguese LLM Leaderboard (archived)+1
751Granite 3.1 1B-A400M Instruct · 8-bit IBM29±2†local~3GB131kOllama › ibm-granite/granite-3.1-3b-a800m-instruct model card
759StarCoderBase 3B · 4-bit BigCode28±1local~3.4GBOllama › 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
760Granite 3.0 1B-A400M Instruct · 8-bit IBM27±1local~3GB4kllama.cpp › IBM Granite 3.0 Language Models technical report (GitHub)EuroEval leaderboards (results collection), English generative leaderboard
763Granite 4.0 H 350M · 4-bit IBM27±1†local~1.9GB33kLM Studio › ibm-granite/granite-4.0-h-1b model card
764Flan-T5 Large · 4-bit Google27±2local~4.8GBhow to run › Scaling Instruction-Finetuned Language Models — authors' own runsVectara hallucination leaderboard (vectara/hallucination-leaderboard repository)
766Gemma 2B IT · 8-bit Google27±1local~5.2GB8kOllama › EvalPlus / EvalPerf published evaluation runsGranite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runsWildBench v2 leaderboard data (Ai2)Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)LMArena leaderboard datasetAlpacaEval 1.0 and 2.0 leaderboards (tatsu-lab/alpaca_eval repository)+3
767EuroLLM 1.7B Instruct · 4-bit UTTER (EuroLLM)27±3local~2.6GB4kllama.cpp › SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)Open Portuguese LLM Leaderboard (archived)EuroEval leaderboards (results collection), English generative leaderboard
768Qwen2.5 0.5B Instruct · 8-bit Alibaba27±2local~2.6GB33kOllama › HuggingFaceTB/SmolLM2-360M-Instruct model cardSpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)Berkeley Function Calling Leaderboard result archiveVectara hallucination leaderboard (vectara/hallucination-leaderboard repository)+1
772Granite 4.0 350M · 16-bit IBM26±1local~2.9GB32kOllama › ibm-granite/granite-4.0-h-1b model cardBerkeley Function Calling Leaderboard result archiveEpoch AI — internal evaluations
776Qwen2 0.5B Instruct · 8-bit Alibaba25±2local~2.6GB32kOllama › Qwen2-0.5B-Instruct model cardLiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasetsOpen Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)Open Portuguese LLM Leaderboard (archived)
778RedPajama-INCITE Base 3B v1 · 4-bit Together25±3local~3.4GB2kllama.cpp › RedPajama — original zero-shot 3B comparisonOpen Portuguese LLM Leaderboard (archived)
781OpenLLaMA 3B v2 · 4-bit OpenLM Research25±1local~4.2GBllama.cpp › Stability AI — original StableLM zero-shot evaluation outputsOpen Portuguese LLM Leaderboard (archived)
784Falcon3 1B Instruct · 8-bit TII24±4local~4.1GB8kOllama › Berkeley Function Calling Leaderboard result archiveCodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)Open Portuguese LLM Leaderboard (archived)
786Qwen1.5 1.8B Chat · 4-bit Alibaba24±2local~3.1GB33kLM Studio › LiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasetsAlpacaEval 1.0 and 2.0 leaderboards (tatsu-lab/alpaca_eval repository)EuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
789Pythia 2.8B · 8-bit EleutherAI21±1local~4.7GB2kllama.cpp › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsRedPajama — original zero-shot 3B comparisonOpen Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)Open Portuguese LLM Leaderboard (archived)
791OLMo 1B · 8-bit Allen Institute for AI21±2†local~2.9GB2kllama.cpp › OLMo: Accelerating the Science of Language Models — authors' own runs
792Phi-1 (1.3B) · 4-bit Microsoft21±4local~4.5GB2khow to run › CRUXEval published evaluation results (authors' generations and scores)Open Portuguese LLM Leaderboard (archived)
793Qwen1.5 0.5B Chat · 4-bit Alibaba21±3local~1.9GB33kllama.cpp › LiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasetsEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
795Pythia 2.8B (deduped) · 8-bit EleutherAI20±1local~4.7GB2kllama.cpp › Stability AI — original StableLM zero-shot evaluation outputsOpen Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)Open Portuguese LLM Leaderboard (archived)
796Flan-T5 Base · 4-bit Google20±3†local~2.5GBhow to run › Scaling Instruction-Finetuned Language Models — authors' own runs
797Falcon-RW 1B · 16-bit TII20±2local~4.3GB2khow to run › Textbooks Are All You Need II: phi-1.5 technical report — authors' own runsHallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)
799SmolLM2 360M Instruct · 4-bit Hugging Face19±3local~2.3GB8kOllama › HuggingFaceTB/SmolLM2-360M-Instruct model cardEuroEval leaderboards (results collection), English generative leaderboard
802Pythia 1.4B · 4-bit EleutherAI18±2local~4.6GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOpen Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)
803BLOOM 3B · 4-bit BigScience18±1local~4GB2kllama.cpp › Stability AI — original StableLM zero-shot evaluation outputsPythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsBLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runsHallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)Open Portuguese LLM Leaderboard (archived)
806OPT-1.3B · 4-bit Meta16±1local~4.3GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsTextbooks Are All You Need II: phi-1.5 technical report — authors' own runsOpen Portuguese LLM Leaderboard (archived)
807Gemma 3 270M IT · 8-bit Google15±4local~2.4GB32kOllama › Google DeepMind Gemma 3 270M IT model cardEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
808Pythia 1B · 4-bit EleutherAI14±1local~3.7GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOLMo: Accelerating the Science of Language Models — authors' own runsOpen Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)Open Portuguese LLM Leaderboard (archived)
810BLOOM 1.7B · 4-bit BigScience14±1local~5.1GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsBLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runsHallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)Open Portuguese LLM Leaderboard (archived)
811Flan-T5 Small · 4-bit Google14±3†local~1.8GBhow to run › Scaling Instruction-Finetuned Language Models — authors' own runs
813GPT-2 XL (1.5B) · 16-bit OpenAI11±1local~4.8GB1kLM Studio › Surface Form Competition (Holtzman et al., EMNLP 2021) — authors' own GPT-2 runsTextbooks Are All You Need II: phi-1.5 technical report — authors' own runsEleutherAI GPT-J-6B model card — EleutherAI lm-evaluation-harness runsPROST (Aroca-Ouellette et al., Findings of ACL-IJCNLP 2021) - authors' own zero-shot runsEleutherAI GPT-Neo 2.7B model card — EleutherAI evaluation-harness runsOpen Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)+2
814BLOOM 1.1B · 4-bit BigScience11±1local~3.7GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsBLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runsHallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)
815Pythia 410M · 4-bit EleutherAI10±2local~2.5GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOpen Portuguese LLM Leaderboard (archived)
818Zyte-1B · 4-bit venkycs (aihub-app)8±2local~3.8GBhow to run › EvalPlus / EvalPerf published evaluation runs
819GPT-Neo 125M · 4-bit EleutherAI5±5local~2.1GB2khow to run › Hallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)COPEN — EMNLP 2022 original zero-shot evaluationsOpen Portuguese LLM Leaderboard (archived)
820BLOOM 560M · 4-bit BigScience5±2local~2.7GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsBLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runsHallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)Open Portuguese LLM Leaderboard (archived)
821OPT-350M · 4-bit Meta5±2local~2.2GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOpen Portuguese LLM Leaderboard (archived)
823Pythia 160M · 4-bit EleutherAI3±2local~1.9GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOpen Portuguese LLM Leaderboard (archived)
824OPT-125M · 4-bit Meta3±2local~1.8GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOpen Portuguese LLM Leaderboard (archived)
825GPT-2 Large (774M) · 4-bit OpenAI1±2local~4.9GB1khow to run › Surface Form Competition (Holtzman et al., EMNLP 2021) — authors' own GPT-2 runsPROST (Aroca-Ouellette et al., Findings of ACL-IJCNLP 2021) - authors' own zero-shot runsCOPEN — EMNLP 2022 original zero-shot evaluationsEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
826GPT-2 Medium (355M) · 4-bit OpenAI-1±2local~3.1GB1khow to run › Surface Form Competition (Holtzman et al., EMNLP 2021) — authors' own GPT-2 runsPROST (Aroca-Ouellette et al., Findings of ACL-IJCNLP 2021) - authors' own zero-shot runsCOPEN — EMNLP 2022 original zero-shot evaluationsEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)
827Pythia 70M · 4-bit EleutherAI-7±2local~1.7GB2khow to run › Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runsOpen Portuguese LLM Leaderboard (archived)
828GPT-2 Small (124M) · 4-bit OpenAI-9±3local~2.1GB1khow to run › Surface Form Competition (Holtzman et al., EMNLP 2021) — authors' own GPT-2 runsPROST (Aroca-Ouellette et al., Findings of ACL-IJCNLP 2021) - authors' own zero-shot runsCOPEN — EMNLP 2022 original zero-shot evaluationsEuroEval leaderboards (results collection), English generative leaderboardOpen Portuguese LLM Leaderboard (archived)

+ 694 more don't fit your machine or budget · show them

+ 32 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them

830 of 1134 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