Best AI models for agents for a 64GB graphics card
A leaderboard of the best AI models for agents for a 64GB graphics card (with 32GB of system RAM). The best one that runs on it right now is Qwen3.8-Flash-Next, followed by Qwen3.8 27B and Ornith 1.5 35B-A3B. Models are ranked on agent tests like SWE-bench and Terminal-Bench, where they use tools, write code and work in a terminal. Models that run on it show the download that fits and the command to run it.
300 models · scores combine 76 benchmarks · updated how it works · find a leaderboard · benchmark your machine
Scores use reviewed agent tasks evaluations only. Compare scores within this task. Models without task measurements are excluded.
Models for 32GB RAM · 64GB graphics card → ~64GB fast · reset
+ more
$ terminal
↗ share
detecting your OS…
Score vs API price
x
show
● open weights ◆ API only ─┘ best score for the price · x: USD per 1M tokens, 3 input : 1 output (log)
| # | model | task score↓ | where↓ | speed↓ | sources |
|---|
| # | model | task score↓ | where↓ | speed↓ | sources |
|---|---|---|---|---|---|
| 1 |  Claude Opus 5.5 | 88±1 | api $8.00/M |     | |
| 2 | Claude Sonnet 5.5 | 86±1 | api $4.00/M | | |
| 3 | Claude Fable 5.1 | 85±1 | api $20.00/M | | |
| 4 |  GPT-6.1 Sol | 85±1* | api $4.00/M | | |
| 5 | GPT-6 Astra | 84±1 | api $20.00/M |  | |
| 6 | Claude Opus 5 | 83±1 | api $10.00/M | +3 | |
| 7 | GPT-6 Sol | 83±1 | api $4.00/M | | |
| 8 | Claude Fable 5 | 82±1 | api $20.00/M | | |
| 9 |  Gemini 4 Argon (limited preview) | 81±2* | limited preview |  | |
| 10 | Claude Mythos Preview (limited preview) | 81±2† | limited preview | | |
| 11 | GPT-5.6 Sol | 80±1 | api $8.00/M | +4 | |
| 12 |  DeepSeek V4.1 Flash | 79±1 | api $0.26/M |  | |
| 13 | Claude Haiku 5.5 | 79±1† | api $0.20/M | | |
| 14 |  GLM-5.3 | 79±1* | api $2.15/M | | |
| 15 | GPT-5.5 Pro | 79±3*† | api $67.50/M | | |
| 16 | GPT-5.4 Pro | 79±3*† | api $67.50/M | | |
| 17 |  MiMo-V2.6-Pro | 78±2 | api $0.54/M | | |
| 18 |  Qwen3.8 Max (2.4T-A95B) | 77±2 | api $3.00/M | | |
| 19 | Claude Opus 4.8 | 77±1 | api $10.00/M | +3 | |
| 20 |  Grok 4.7 | 77±2* | api $3.00/M | | |
| 21 |  Hy4 preview (770B-A49B) | 77±2 | api $1.25/M | | |
| 22 | Qwen3.8 Max (0902) | 76±2* | api $3.00/M | | |
| 23 |  Kimi K3 | 76±1 | api $6.00/M | | |
| 24 | GLM-5.3-Flash | 76±1 | api $0.24/M | | |
| 25 | GPT-5.5 | 76±1 | api $11.25/M | +6 | |
| 26 | GPT-5.6 Terra | 76±1 | api $4.50/M | | |
| 27 | Gemini 3.8 Flash | 75±1 | api $1.50/M |  +1 | |
| 28 | MiMo-V2.6-Flash (309B-A15B) | 75±2 | api $0.17/M | | |
| 29 |  Muse Spark 1.3 | 75±1 | api $2.00/M | | |
| 30 | Grok 4.6 | 74±1 | api $3.00/M | | |
| 31 | DeepSeek V4 Pro (0813) | 74±1 | api $1.63/M | | |
| 33 | GPT-5.4 | 74±1 | api $5.63/M | +4 | |
| 34 | Claude Opus 4.7 | 74±2 | api $10.00/M |  +1 | |
| 35 | GPT-6 Luna | 74±1 | api $0.20/M | | |
| 36 | Claude Sonnet 5 | 74±1 | api $4.00/M | | |
| 37 | GPT-5.6 Luna | 74±1 | api $0.45/M | | |
| 38 | Grok 4.5 | 73±1 | api $3.00/M | | |
| 40 | GLM-5.2 | 73±1 | api $2.15/M | +3 | |
| 41 | Qwen3.8-Flash-Next · 2-bit | 72±2 | local |  | |
| 42 | Gemini 3.7 Flash | 72±1 | api $1.50/M | | |
| 43 | Qwen3.8-Omni-Flash | 72±3† | api $0.23/M | | |
| 44 | DeepSeek V4 Flash (0731) | 71±2 | api $0.11/M | | |
| 45 | Claude Sonnet 4.6 | 71±1 | api $6.00/M | +1 | |
| 46 | Qwen3.8 27B · 8-bit | 71±1 | local | | |
| 47 | GPT-5.3-Codex | 71±2 | api $4.81/M | | |
| 48 | Claude Opus 4.6 | 70±1 | api $10.00/M | +6 | |
| 49 | Hunyuan Hy3 (295B-A21B) | 70±1 | api $0.23/M | | |
| 50 | Gemini 3.5 Flash | 70±3 | api $3.38/M | | |
| 51 | Muse Spark 1.2 | 69±2* | api $2.00/M | | |
| 52 | Muse Spark 1.1 | 69±2* | api $2.00/M | | |
| 53 | Kimi K2.6 | 68±1 | api $1.71/M | +2 | |
| 54 | Qwen3.7 Max | 68±2 | api $3.75/M | | |
| 55 | Claude Opus 4.5 | 68±1 | api $10.00/M | +4 | |
| 56 | GLM-5.1 | 67±1 | api $2.15/M | +5 | |
| 58 |  MiniMax-M3 | 67±2 | api $0.48/M | | |
| 59 | GPT-5.2 | 67±1 | api $4.81/M | +6 | |
| 60 | Gemini 3.1 Pro (Preview) | 66±1 | api $4.50/M | +4 | |
| 61 | Qwen3.7 Plus | 66±2 | api $0.70/M | | |
| 62 | GPT-5.4 mini | 66±2 | api $1.69/M | | |
| 64 |  Mistral Large 4 (public preview) | 66±2* | api $2.06/M | | |
| 66 | MiMo-V2.5 (310B-A15B) | 65±2 | api $0.21/M | | |
| 68 | Qwen3.5 397B-A17B | 65±1 | api $1.09/M +local | | |
| 69 |  Ornith 1.5 35B-A3B · 8-bit | 65±2† | local | | |
| 70 | Qwen3.6 Max Preview | 65±5* | api $2.92/M | | |
| 72 | Muse Glimmer 30B · 8-bit | 65±2 | local | | |
| 73 | Gemini 3.6 Flash | 64±2 | api $1.50/M | | |
| 74 |  LongCat 2.0 (1.6T-A48B) | 64±6* | api $1.30/M | | |
| 75 | GLM-5 | 64±1 | api $1.55/M | +1 | |
| 76 | Qwen3.6 27B · 8-bit | 64±1 | local | | |
| 77 | GPT-5 Pro | 64±5*† | api $41.25/M | | |
| 78 | MiMo-V2.5-Pro (1T-A42B) | 64±1 | api $0.65/M | | |
| 80 |  Inkling Small (266B-A12B) | 64±2 | api $0.64/M +local | | |
| 81 | MiniMax-M2.7 | 64±2 | api $0.53/M +local | | |
| 82 | Gemini 3 Flash (Preview) | 64±2 | api $1.13/M | | |
| 83 | Inkling (975B-A41B) | 63±2 | api $1.73/M | | |
| 84 |  Step 3.5 Flash (196B-A11B) | 63±2† | api $0.15/M | | |
| 85 | GPT-5 | 63±2 | api $3.44/M | +2 | |
| 87 |  Nemotron 3 Ultra 550B-A55B | 63±2† | api $1.35/M | | |
| 88 |  KAT-Coder-V2.5-Dev (35B-A3B) · 8-bit | 62±2† | local | | |
| 89 | Qwen3.5 27B · 8-bit | 62±1 | local | | |
| 90 | Qwen3.5 122B-A10B | 62±2 | api $1.10/M +local | | |
| 92 | Kimi K2.5 | 62±1 | api $1.20/M | +2 | |
| 93 | Qwen3.6 35B-A3B · 8-bit | 61±1 | local | | |
| 94 | Claude Sonnet 4.5 | 61±1 | api $6.00/M | +3 | |
| 95 | GLM-4.7 · 1-bit | 61±1† | local | | |
| 96 | GPT-5.4 nano | 61±2 | api $0.46/M | | |
| 97 | GPT-5.1 | 61±2 | api $3.44/M | | |
| 98 | Ornith 1.0 35B (35B-A3B) · 8-bit | 60±1 | local | | |
| 99 |  Nova 2 Pro (Preview) | 60±8*† | api $3.78/M | | |
| 100 | Qwen3.5 35B-A3B · 8-bit | 60±1 | local | | |
| 101 | MiniMax-M2.1 · 2-bit | 59±2 | local (slow) | | |
| 103 |  SmolLM3 3B · 8-bit | 59±11*† | local | | |
| 105 | DeepSeek V3.2 | 59±2 | api $0.29/M | | |
| 106 | MiniMax-M2.5 | 59±2 | api $0.53/M +local | | |
| 109 | Mistral Medium 3.5 (128B) · 2-bit | 58±2 | local | | |
| 110 | MiniMax-M2 | 58±2 | api $0.53/M +local | | |
| 111 | Qwen3.5 Plus | 58±4* | api $0.90/M | | |
| 112 | Grok 4.20 (0309 reasoning) | 58±6* | api $1.56/M | | |
| 114 |  Gemma 4 31B · 8-bit | 57±1 | local | +4 | |
| 115 | Grok 4.3 | 57±3 | api $1.56/M | | |
| 116 | GLM-4.6 · 1-bit | 57±1 | local | | |
| 117 | Grok 4.1 Fast (Reasoning) | 57±2 | api $0.28/M | | |
| 119 |  Spark-X2.5 4B · 16-bit | 56±2† | local | | |
| 120 | Qwen3.6 Plus | 56±4* | api $1.13/M | | |
| 121 | Qwen3-Max (2025-09-23) | 56±3† | api $2.40/M | | |
| 122 | Kimi K2 Thinking | 56±2 | api $1.07/M | | |
| 123 |  EXAONE 4.5 33B · 8-bit | 56±3*† | local | | |
| 124 | Kimi K2 Instruct (0711) | 56±1 | api $1.00/M | | |
| 125 |  Solar Pro 4 | 56±2* | api $0.53/M | | |
| 127 | MiMo-V2.6-Distill-Qwen-9B · 8-bit | 56±3† | local | | |
| 129 | o3-pro | 56±5* | api $35.00/M | | |
| 130 | GPT-5 mini | 55±1 | api $0.69/M | +5 | |
| 134 | Claude Haiku 4.5 | 55±2 | api $2.00/M | +1 | |
| 136 | GLM-4.5 · 1-bit | 55±2 | local | | |
| 137 |  MiniCPM5 2B · 8-bit | 55±3† | local | | |
| 139 | Nemotron 3 Super 120B-A12B | 54±2† | api $0.28/M +local | | |
| 140 | GPT-4o (2024-05-13) | 54±11* | api $7.50/M | | |
| 141 | Gemini 3.1 Flash-Lite | 54±6* | api $0.56/M | | |
| 142 |  Granite 4.2 30B · 8-bit | 54±2† | local | | |
| 144 | Qwen3-Coder 480B-A35B Instruct | 53±2 | api $0.67/M +local | | |
| 146 | o3 | 53±1 | api $3.50/M | +1 | |
| 147 | Gemini 2.5 Pro | 53±2 | api $3.44/M | +3 | |
| 148 | Nemotron-Labs-3 Puzzle 75B-A9B · 4-bit | 53±2† | local | | |
| 149 |  Seed-OSS 36B Instruct · 8-bit | 52±3† | local | | |
| 151 | Mistral Small 4 (119B-A6.5B) · 2-bit | 52±3* | local | | |
| 153 |  Command A+ (05-2026) | 52±2 | api price unknown +local |  | |
| 155 | GLM-4.5-Air · 2-bit | 51±2 | local | | |
| 156 |  Jamba 1.5 Mini · 4-bit | 51±10*† | local | | |
| 157 | Qwen3.5 9B · 8-bit | 51±1 | local | +3 | |
| 158 | DeepSeek V3.1 | 51±1 | api $0.42/M | | |
| 159 |  Mellum2.1 12B-A2.5B Thinking · 16-bit | 50±3† | local | | |
| 160 | Qwen3 235B-A22B Thinking 2507 | 50±1 | api $0.97/M +local | | |
| 161 | gpt-oss-120b | 49±2 | api $0.26/M +local | +1 | |
| 162 | Qwen3-Next 80B-A3B Thinking · 4-bit | 49±2† | local | | |
| 163 | Command A Reasoning (08-2025) · 2-bit | 49±3 | local | | |
| 164 | Granite 4.1 30B · 4-bit | 49±5*† | local | | |
| 165 | o4-mini | 49±1 | api $1.93/M | | |
| 166 | Gemma 4 12B · 8-bit | 48±2 | local | | |
| 168 | Gemma 4 26B-A4B · 8-bit | 48±1 | local | +1 | |
| 169 | GPT-4.1 | 48±1 | api $3.50/M | +4 | |
| 170 | Qwen3-VL 32B Thinking · 8-bit | 48±1 | local | | |
| 171 | Granite 4.2 8B · 8-bit | 48±2† | local | | |
| 172 | Nova Premier | 47±3*† | api $5.00/M |  | |
| 174 | Qwen3.5 4B · 8-bit | 47±1 | local | +1 | |
| 175 | Qwen3 235B-A22B Instruct 2507 | 47±1 | api $0.20/M +local | +1 | |
| 177 | Qwen3-VL 32B Instruct · 8-bit | 46±3*† | local | | |
| 180 | Solar Open 100B (102B-A12B) · 2-bit | 46±3† | local | | |
| 181 | GPT-4o (2024-11-20) | 46±1 | api $4.38/M | +1 | |
| 182 |  LFM2.5 2.6B · 8-bit | 46±2 | local | | |
| 184 | GPT-4.1 mini | 46±1 | api $0.70/M | +1 | |
| 185 | Qwen3 4B Thinking 2507 · 8-bit | 46±2† | local | | |
| 187 | GPT-4 Turbo (2024-04-09) | 46±1* | api $15.00/M | | |
| 188 | Nemotron-Cascade-14B-Thinking · 8-bit | 46±4*† | local | | |
| 189 | Qwen3 14B · 8-bit | 46±1 | local | | |
| 191 | Qwen3 Coder 30B-A3B Instruct · 8-bit | 45±2 | local | | |
| 192 | GPT-5 nano | 45±2 | api $0.14/M | | |
| 194 | Qwen3 30B-A3B Thinking 2507 · 8-bit | 45±1 | local | | |
| 195 | Nemotron 3 Nano 30B-A3B · 8-bit | 45±3*† | local | | |
| 196 | Granite 4.2 3B · 8-bit | 45±2 | local | | |
| 197 | Mellum2 12B-A2.5B Instruct · 16-bit | 45±4*† | local | | |
| 198 | Nemotron Nano 9B v2 · 8-bit | 45±4*† | local | | |
| 200 | Qwen3-VL 4B Thinking · 8-bit | 45±3*† | local | | |
| 201 | Gemini 2.5 Flash | 45±1 | api $0.85/M | | |
| 202 | EXAONE 4.0 32B · 8-bit | 45±1 | local | | |
| 203 | Qwen3-VL 30B-A3B Thinking · 8-bit | 44±2*† | local | | |
| 204 | Qwen3-VL 30B-A3B Instruct · 8-bit | 44±3*† | local | | |
| 205 | Qwen3 32B · 8-bit | 44±1 | local | +1 | |
| 206 | Command A (03-2025) · 2-bit | 44±1 | local | | |
| 207 |  Falcon-H1R 7B · 8-bit | 44±2*† | local | | |
| 208 | Qwen3 235B-A22B (Apr 2025, hybrid) | 44±1 | api $1.23/M +local | +1 | |
| 209 | Qwen3 8B · 8-bit | 44±1 | local | +1 | |
| 210 | GPT-4o (2024-08-06) | 44±2 | api $4.38/M | | |
| 211 | Gemini 2.5 Flash-Lite | 43±3* | api $0.17/M | | |
| 212 | Qwen3-Next 80B-A3B Instruct · 4-bit | 43±2† | local | | |
| 213 | Qwen3-VL 8B Instruct · 8-bit | 43±3* | local | | |
| 214 | DeepSeek R1 0528 | 43±1 | api $0.91/M | | |
| 215 | Nemotron-Cascade-8B · 8-bit | 43±4*† | local | | |
| 217 |  Apriel 1.6 15B Thinker · 8-bit | 43±4† | local | | |
| 218 | Spark-X2.5 1.7B · 16-bit | 43±2† | local | | |
| 219 | Qwen2.5 72B Instruct · 4-bit | 43±1 | local | | |
| 220 | Qwen2.5 32B Instruct · 8-bit | 43±1* | local | | |
| 222 | Qwen3-VL 4B Instruct · 8-bit | 42±3*† | local | | |
| 223 | Qwen3 30B-A3B Instruct 2507 · 8-bit | 42±1 | local | | |
| 226 | Qwen3-VL 8B Thinking · 8-bit | 42±3*† | local | | |
| 227 | gpt-oss-20b · MXFP4 | 42±2 | local | | |
| 229 | Llama 3.1 405B Instruct | 42±2 | api $2.40/M | | |
| 230 | Ministral 3 14B · 8-bit | 42±3 | local | | |
| 231 | Qwen3 30B-A3B (Apr 2025, hybrid) · 8-bit | 42±1 | local | | |
| 233 | Granite 4.1 8B · 8-bit | 41±2*† | local | | |
| 234 | GPT-4o mini (2024-07-18) | 41±1 | api $0.26/M | | |
| 235 | Granite 4.1 3B · 4-bit | 41±5*† | local | | |
| 237 | Mistral Large 3 (675B-A41B) | 41±8* | api $0.75/M | | |
| 239 | Ministral 3 8B · 8-bit | 41±4* | local | | |
| 240 | Mistral Small 3.2 24B · 8-bit | 41±3* | local | | |
| 241 | Nova Pro | 41±1 | api $1.40/M | | |
| 242 | LFM2.5 8B-A1B · 8-bit | 41±2 | local | | |
| 243 | Qwen3 4B · 8-bit | 41±1 | local | | |
| 244 | Qwen2.5 14B Instruct · 8-bit | 40±1* | local | | |
| 245 | LFM2.5 1.2B Thinking · 8-bit | 40±4*† | local | | |
| 247 | DeepSeek V3 (Dec '24) | 40±1 | api $0.46/M | | |
| 248 | DeepSeek R1 (Jan '25) | 40±1 | api $4.00/M | | |
| 249 | Mistral Small 3.1 24B (2503) · 8-bit | 40±1 | local | | |
| 250 | Mistral Large 2 (2411) · 4-bit | 40±1 | local (slow) | | |
| 251 | Mistral Large 2 (2407) · 2-bit | 40±1* | local | | |
| 252 | QwQ-32B · 8-bit | 40±1* | local | | |
| 253 | Qwen3 4B Instruct 2507 · 8-bit | 40±1 | local | | |
| 254 | o1 (2024-12-17) | 40±1 | api $26.25/M | | |
| 257 | Llama 3.1 Nemotron Ultra 253B v1 · 2-bit | 39±1* | local (slow) | | |
| 259 | Gemma 4 E4B · 8-bit | 39±1 | local | +3 | |
| 260 |  Phi-3.5-mini-instruct · 8-bit | 39±11*† | local | | |
| 261 | Qwen3-VL 2B Thinking · 8-bit | 39±3*† | local | | |
| 262 | Nova 2 Lite | 39±3* | api $0.94/M | | |
| 263 | Nova Lite | 39±1 | api $0.10/M | | |
| 264 | Qwen2.5 7B Instruct · 8-bit | 38±1 | local | | |
| 265 | Granite 4.0 1B · 16-bit | 38±3* | local | | |
| 268 | Qwen3 1.7B · 8-bit | 38±1 | local | | |
| 269 | Qwen3-VL 2B Instruct · 8-bit | 38±3*† | local | | |
| 271 | Gemma 3 27B · 8-bit | 38±1 | local | | |
| 272 | Ministral 3 14B Reasoning 2512 · 4-bit | 38±5* | local | | |
| 273 | o3-mini | 37±1 | api $1.93/M | | |
| 274 | Llama 4 Maverick (402B-A17B) | 37±1 | api $0.42/M +local | | |
| 275 | GPT-4.1 nano | 37±1 | api $0.17/M | | |
| 277 | Llama 3.1 70B Instruct · 4-bit | 37±1 | local | | |
| 278 | Nemotron 3 Nano 4B · 8-bit | 37±2 | local | | |
| 279 | GPT-3.5 Turbo (0125) | 37±1* | api $0.75/M | | |
| 280 |  Sky-T1 32B Preview · 4-bit | 37±1* | local | | |
| 281 | Granite 4.0 H 1B · 8-bit | 37±3* | local | | |
| 282 | Llama 3.3 70B Instruct · 4-bit | 36±1 | local | | |
| 284 | LFM2.5 1.2B Instruct · 4-bit | 36±4*† | local | | |
| 285 | Command R7B (12-2024) · 8-bit | 36±1 | local | | |
| 286 | Mellum2 12B-A2.5B Thinking · 16-bit | 36±3† | local | | |
| 287 | Gemma 3 12B · 8-bit | 36±1* | local | | |
| 288 | Llama 3 70B Instruct · 4-bit | 36±1* | local | | |
| 289 | Nova Micro | 36±1 | api $0.061/M | | |
| 290 | Ministral 8B (2410) · 16-bit | 36±1* | local | | |
| 291 | Qwen2.5 3B Instruct · 8-bit | 36±1* | local | | |
| 292 | Mistral Small Instruct 22B (2409) · 8-bit | 36±4* | local | | |
| 293 | Mistral NeMo Instruct (2407) · 8-bit | 35±1 | local | | |
| 294 | Granite 3.1 8B Instruct · 8-bit | 35±4* | local | | |
| 295 | MiniCPM3-4B · 4-bit | 35±1* | local | | |
| 296 | Granite 3.2 8B Instruct · 8-bit | 35±4* | local | | |
| 297 | Llama 3.1 8B Instruct · 8-bit | 35±1 | local | | |
| 298 | Command R Plus (04-2024) · 2-bit | 34±1* | local | | |
| 299 | DeepSeek R1 0528 Qwen3 8B · 8-bit | 34±5* | local | | |
| 300 | Qwen2.5 1.5B Instruct · 8-bit | 34±1* | local | | |
| 301 | Phi-4 14B · 8-bit | 34±1* | local | | |
| 302 | Command R Plus (08-2024) | 34±7*† | api $4.38/M +local | | |
| 303 | Qwen3 0.6B · 8-bit | 34±1* | local | | |
| 304 |  Hermes 2 Pro Llama 3 8B · 4-bit | 34±1* | local | | |
| 305 | Granite 4.0 H 350M · 8-bit | 34±5*† | local | | |
| 307 | EXAONE 4.0 1.2B · 8-bit | 33±3† | local | | |
| 308 | Falcon3 10B Instruct · 8-bit | 33±1* | local | | |
| 309 | Llama 4 Scout (109B-A17B) · 2-bit | 33±1 | local | | |
| 310 | Llama 3.2 3B Instruct · 8-bit | 33±1 | local | | |
| 312 | Gemma 4 E2B · 8-bit | 32±2 | local | +2 | |
| 313 | Falcon3 7B Instruct · 8-bit | 32±1* | local | | |
| 314 | Qwen3.5 2B · 8-bit | 32±2 | local | | |
| 315 | Magistral Small 1.0 (2506) · 8-bit | 31±7* | local | | |
| 316 | Mixtral 8x7B Instruct v0.1 · 8-bit | 31±1* | local | | |
| 317 | Mixtral 8x22B Instruct v0.1 · 2-bit | 31±1* | local | | |
| 318 | Qwen2 7B Instruct · 8-bit | 31±1* | local | | |
| 319 |  Bielik 11B v2.3 Instruct · 8-bit | 30±1* | local | | |
| 320 | Command R (08-2024) · 8-bit | 30±8*† | local | | |
| 321 | Ministral 3 3B · 8-bit | 30±6* | local | | |
| 322 |  ChatGLM3-6B · 4-bit | 30±10*† | local | | |
| 323 | GLM-4 9B Chat · 8-bit | 30±1 | local | | |
| 324 | Gemma 3 4B · 8-bit | 30±1* | local | | |
| 325 |  DBRX Instruct · 4-bit | 30±1* | local (slow) | | |
| 326 | Granite 4.0 350M · 16-bit | 30±3* | local | | |
| 327 |  Olmo 3 7B Instruct · 8-bit | 29±4 | local | | |
| 329 |  NexusRaven V2 13B · 8-bit | 28±1* | local | | |
| 330 | Qwen2.5 0.5B Instruct · 8-bit | 28±1* | local | | |
| 331 | Qwen3.5 0.8B · 8-bit | 27±5*† | local | | |
| 332 | Llama 3 8B Instruct · 8-bit | 26±1* | local | | |
| 333 | Falcon3 3B Instruct · 8-bit | 26±1* | local | | |
| 334 | Phi-4-mini 3.8B · 8-bit | 23±1 | local | | |
| 335 | Qwen2 1.5B Instruct · 8-bit | 23±2* | local | | |
| 336 | Llama 3.2 1B Instruct · 4-bit | 18±1 | local | | |
| 337 | DeepSeek-Coder-V2 Lite Instruct (16B-A2.4B) · 8-bit | 18±2* | local | | |
| 338 | QwQ-32B-Preview · 8-bit | 17±2* | local | | |
| 339 | Falcon3 1B Instruct · 8-bit | 17±2* | local | | |
| 340 | Gemma 3 1B · 4-bit | 13±2 | local | | |
| — | Devstral 2 (2512) · q4_k_m | not yet calibrated | local (slow) | | |
| — | Devstral Small 2 24B · q8_0 | not yet calibrated | local | | |
| — | GLM-4.7-Flash · Q8_0 | not scored yet | local | ||
| — | Granite 4.0 H Micro 3B · q4_k_m | not yet calibrated | local | | |
| — | Granite 4.0 H Small (32B-A9B) · Q8_0 | not yet calibrated | local | | |
| — | Granite 4.0 H Tiny (7B-A1B) · q4_k_m | not yet calibrated | local | | |
| — | Granite 4.0 Micro 3B · Q8_0 | not yet calibrated | local | | |
| — |  K2 Horizon 7B · gguf-q8_0 | not scored yet | local | ||
| — |  Laguna M.1 | not yet calibrated | api price unknown | | |
| — | Laguna XS 2.1 (33B-A3B) · q8_0 | not yet calibrated | local | | |
| — | Laguna XS.2 · q4_k_m | not yet calibrated | local | | |
| — |  Ling 3.0 Flash (124B-A5B) | not yet calibrated | api $0.090/M +local | | |
| — | Llama 3.3 Nemotron Super 49B v1 · q4_k_m | not scored yet | local | ||
| — | MiniMax-M1 (80k) | not yet calibrated | api $0.96/M | | |
| — | Nemotron Cascade 2 30B-A3B · Q8_0 | not scored yet | local | ||
| — | Qwen3-Coder-Next (80B-A3B) · q4_k_m | not scored yet | local | ||
| — | Step 3.7 Flash (198B-A11B) | not scored yet | api $0.44/M | ||
| — |  Trinity-Large-Thinking | not scored yet | api $0.39/M |
+ 62 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
362 of 398 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