Best AI models for math for a 32GB graphics card
A leaderboard of the best AI models for math for a 32GB graphics card (with 32GB of system RAM). The best one that runs on it right now is INTELLECT-3, followed by Qwen3-Coder-Next (80B-A3B) and EXAONE 4.5 33B. Models are ranked on math tests, from competition problems like AIME to research problems like FrontierMath. Models that run on it show the download that fits and the command to run it.
406 models · scores combine 59 benchmarks · updated how it works · find a leaderboard · benchmark your machine
Scores use reviewed math evaluations only. Compare scores within this task. Models without task measurements are excluded.
Models for 32GB RAM · 32GB graphics card → ~32GB 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 |  GPT-6 Astra | 92±1 | api $20.00/M |   | |
| 2 | GPT-6.1 Sol | 90±2 | api $4.00/M | | |
| 3 |  Claude Opus 5.5 | 86±1 | api $8.00/M | | |
| 4 | Claude Fable 5.1 | 84±1 | api $20.00/M | | |
| 5 | GPT-6 Sol | 84±1 | api $4.00/M | | |
| 6 | Claude Fable 5 | 82±1 | api $20.00/M | | |
| 7 | GPT-5.6 Sol | 82±1 | api $8.00/M |  | |
| 8 | Claude Sonnet 5.5 | 81±1 | api $4.00/M | | |
| 9 | Claude Mythos Preview (limited preview) | 81±7*† | limited preview |  | |
| 10 | GPT-5.5 Pro | 81±1* | api $67.50/M | | |
| 11 | GPT-5.5 | 80±1 | api $11.25/M |  | |
| 12 | Claude Opus 5 | 80±1 | api $10.00/M | | |
| 13 | GPT-5.6 Terra | 79±1 | api $4.50/M | | |
| 14 | GPT-5.4 Pro | 77±1 | api $67.50/M | | |
| 15 | GPT-5.6 Luna | 77±1 | api $0.45/M | | |
| 16 | GPT-6 Luna | 76±1 | api $0.20/M | | |
| 17 | Claude Opus 4.8 | 76±1 | api $10.00/M | | |
| 18 | GPT-5.4 | 75±1 | api $5.63/M | | |
| 19 |  Qwen3.8 Max (2.4T-A95B) | 75±1 | api $3.00/M | | |
| 21 | Claude Haiku 5.5 | 74±1 | api $0.20/M | | |
| 22 | GPT-5.2 Pro | 74±1* | api $57.75/M | | |
| 23 |  Muse Spark 1.3 | 74±1 | api $2.00/M | | |
| 24 |  Kimi K3 | 74±1 | api $6.00/M | | |
| 25 |  Gemini 3.7 Flash | 72±1 | api $1.50/M | | |
| 26 | Qwen3.8 Max (0902) | 72±1 | api $3.00/M | | |
| 27 | Qwen3.7 Max | 72±1 | api $3.75/M | | |
| 28 |  Grok 4.6 | 72±1 | api $3.00/M | | |
| 29 |  GLM-5.3 | 72±1 | api $2.15/M | | |
| 30 | Gemini 3.1 Pro (Preview) | 71±1 | api $4.50/M | | |
| 31 | Gemini 3.8 Flash | 71±1 | api $1.50/M | | |
| 32 | GPT-5.2 | 71±1 | api $4.81/M | | |
| 33 |  DeepSeek V4.1 Flash | 71±1 | api $0.26/M | | |
| 34 | DeepSeek V4 Pro (0813) | 71±1 | api $1.63/M | | |
| 35 | Claude Sonnet 5 | 71±1 | api $4.00/M | | |
| 36 | Muse Spark 1.1 | 70±2* | api $2.00/M | | |
| 37 | Grok 4.5 | 70±1 | api $3.00/M | | |
| 38 | Claude Opus 4.6 | 70±1 | api $10.00/M | | |
| 39 | Claude Opus 4.7 | 70±1 | api $10.00/M | | |
| 40 | DeepSeek V4 Flash (0731) | 70±1 | api $0.11/M | | |
| 41 | Gemini 3.5 Flash | 70±1 | api $3.38/M | | |
| 42 | Kimi K2.6 | 70±1 | api $1.71/M | | |
| 43 | GLM-5.2 | 69±1 | api $2.15/M | | |
| 45 | Gemini 3.6 Flash | 69±1 | api $1.50/M | | |
| 46 | GPT-5 | 69±1 | api $3.44/M | | |
| 48 | GPT-5 Pro | 69±1* | api $41.25/M | | |
| 49 | GLM-5.3-Flash | 69±1 | api $0.24/M | | |
| 50 | Grok 4.7 | 68±1 | api $3.00/M | | |
| 53 | Gemini 3 Flash (Preview) | 68±1 | api $1.13/M | | |
| 54 | Kimi K2.7 Code | 68±1 | api $1.71/M | | |
| 55 |  INTELLECT-3 · 2-bit | 67±2† | local (slow) | | |
| 56 | Grok 4.20 (0309 reasoning) | 67±1 | api $1.56/M | | |
| 57 |  Inkling Small (266B-A12B) | 67±1 | api $0.64/M | | |
| 58 | GLM-4.7 | 67±2 | api $1.00/M | | |
| 59 | GPT-5.4 mini | 67±1 | api $1.69/M | | |
| 60 |  LongCat 2.0 (1.6T-A48B) | 67±6*† | api $1.30/M | | |
| 63 | Grok 4.3 | 67±1 | api $1.56/M | | |
| 64 | GPT-5 mini | 66±1 | api $0.69/M | | |
| 65 | GPT-5.4 nano | 66±1 | api $0.46/M | | |
| 66 | Qwen3-Coder-Next (80B-A3B) · 2-bit | 66±5* | local | | |
| 67 | Qwen3.6 Max Preview | 66±4* | api $2.92/M | | |
| 68 | o3-pro | 66±4*† | api $35.00/M | | |
| 69 | Qwen3.6 Plus | 66±1 | api $1.13/M | | |
| 70 | Qwen3 235B-A22B Thinking 2507 | 66±1 | api $0.97/M | | |
| 71 |  EXAONE 4.5 33B · 4-bit | 65±4*† | local | | |
| 72 | Kimi K2.5 | 65±2 | api $1.20/M | | |
| 73 | GLM-5 | 65±1 | api $1.55/M | | |
| 74 |  Granite 4.2 30B · 4-bit | 65±3*† | local | | |
| 77 | GLM-5.1 | 65±1 | api $2.15/M | | |
| 79 | GPT-5.1 | 65±3* | api $3.44/M | | |
| 80 | DeepSeek V3.2 | 65±2 | api $0.29/M | | |
| 81 | Qwen3.5 122B-A10B | 64±2† | api $1.10/M | | |
| 82 |  Step 3.5 Flash (196B-A11B) | 64±1 | api $0.15/M | | |
| 83 | Qwen3.7 Plus | 64±1* | api $0.70/M | | |
| 84 | Qwen3.6 27B · 4-bit | 64±1 | local | | |
| 85 | o4-mini | 64±1 | api $1.93/M |  | |
| 86 | Qwen3.5 Plus | 64±3* | api $0.90/M | | |
| 87 | o3 | 64±1 | api $3.50/M | | |
| 88 | Qwen3.5 397B-A17B | 64±1 | api $1.09/M | +1 | |
| 90 | Gemini 2.5 Flash | 64±2 | api $0.85/M | | |
| 91 | GLM-4.6 | 64±2† | api $1.00/M | | |
| 92 |  Spark-X2.5 4B · 16-bit | 64±2† | local | | |
| 94 | Inkling (975B-A41B) | 64±1 | api $1.73/M | | |
| 95 | Grok 4.1 Fast (Reasoning) | 64±2 | api $0.28/M | | |
| 96 | Qwen3 235B-A22B Instruct 2507 | 63±1 | api $0.20/M | | |
| 97 | DeepSeek V3.1 | 63±1 | api $0.42/M | | |
| 98 | GLM-4.5-Air | 63±2 | api $0.42/M +local | | |
| 99 | Qwen3.5 27B · 4-bit | 63±1 | local | | |
| 100 | Claude Opus 4.5 | 63±1 | api $10.00/M | | |
| 101 |  Nemotron Cascade 2 30B-A3B · 4-bit | 63±2† | local | | |
| 104 |  Gemma 4 26B-A4B · 8-bit | 63±2 | local | | |
| 105 | Qwen3-Next 80B-A3B Instruct · 2-bit | 63±3† | local | | |
| 106 | DeepSeek R1 0528 | 63±1 | api $0.91/M | | |
| 107 | Kimi K2 Thinking | 63±3* | api $1.07/M | | |
| 108 | Step 3.7 Flash (198B-A11B) | 63±1 | api $0.44/M | | |
| 110 | gpt-oss-120b | 62±1 | api $0.26/M | | |
| 111 |  Command A+ (05-2026) | 62±2 | api price unknown |  | |
| 112 | Qwen3-VL 32B Thinking · 4-bit | 62±2 | local |  | |
| 113 | Claude Haiku 4.5 | 62±1* | api $2.00/M | | |
| 114 | Claude Sonnet 4.5 | 62±1 | api $6.00/M | | |
| 115 | Nemotron 3 Ultra 550B-A55B | 62±2 | api $1.35/M | | |
| 117 | Nemotron-Labs-3 Puzzle 75B-A9B · 2-bit | 62±3*† | local (slow) | | |
| 118 | GLM-4.5 | 62±1 | api $1.00/M | | |
| 120 |  Olmo 3.1 32B Think · 4-bit | 62±2† | local | | |
| 121 | Qwen3-Max (2025-09-23) | 62±1 | api $2.40/M | | |
| 122 | Gemini 3.1 Flash-Lite | 62±1 | api $0.56/M | | |
| 123 | Gemma 4 31B · 4-bit | 61±2 | local | | |
| 125 | Nemotron 3 Super 120B-A12B | 61±1 | api $0.28/M | | |
| 126 | Gemini 2.5 Pro | 61±1 | api $3.44/M | | |
| 127 | o3-mini | 61±1 | api $1.93/M | | |
| 128 | Nemotron-Cascade-14B-Thinking · 8-bit | 61±2*† | local | | |
| 130 | Granite 4.2 8B · 8-bit | 61±3*† | local | | |
| 131 |  Falcon-H1R 7B · 8-bit | 61±2† | local | | |
| 132 | Qwen3-Next 80B-A3B Thinking · 2-bit | 61±2† | local | | |
| 133 | Qwen3-VL 30B-A3B Thinking · 4-bit | 61±3*† | local | | |
| 134 | Qwen3.5 35B-A3B · 4-bit | 61±1 | local | | |
| 135 | Qwen3.6 35B-A3B · 4-bit | 61±1 | local | | |
| 137 | GPT-5 nano | 60±1 | api $0.14/M | | |
| 139 | Nemotron-Cascade-8B · 8-bit | 60±2*† | local | | |
| 140 | Qwen3.7 Flash | 60±1* | api $0.055/M | | |
| 141 | Kimi K2 Instruct (0711) | 60±2* | api $1.00/M | | |
| 143 | Olmo 3 32B Think · 4-bit | 60±1† | local | | |
| 144 | o1 (2024-12-17) | 60±1 | api $26.25/M | | |
| 145 | Gemini 3.5 Flash-Lite | 60±1 | api $0.85/M | | |
| 146 | gpt-oss-20b · MXFP4 | 60±1 | local | | |
| 147 |  MiniCPM5 2B · 8-bit | 60±2† | local | | |
| 148 |  MiniMax-M3 | 60±2* | api $0.48/M | | |
| 149 |  Seed-OSS 36B Instruct · 4-bit | 60±2 | local | | |
| 150 | Qwen3 30B-A3B Instruct 2507 · 4-bit | 59±1 | local | | |
| 152 | Qwen3.6 Flash | 59±1* | api $0.56/M | | |
| 153 | Nemotron 3 Nano 30B-A3B · 4-bit | 59±2† | local | | |
| 155 | Qwen3.5 9B · 8-bit | 59±1 | local | | |
| 156 | Qwen3 14B · 8-bit | 59±2* | local | | |
| 157 | QwQ-32B · 4-bit | 58±2 | local | | |
| 158 | Qwen3-VL 8B Thinking · 8-bit | 58±2 | local | | |
| 159 | Claude Sonnet 4.6 | 58±2 | api $6.00/M | | |
| 160 | Qwen3.5 4B · 8-bit | 58±1 | local | | |
| 161 | DeepSeek R1 (Jan '25) | 58±1 | api $4.00/M | | |
| 162 | EXAONE 4.0 32B · 4-bit | 58±2† | local | | |
| 163 | Granite 4.2 3B · 8-bit | 58±1 | local | | |
| 164 | DeepSeek R1 0528 Qwen3 8B · 8-bit | 58±1 | local | | |
| 165 | Qwen3-VL 4B Thinking · 8-bit | 57±2*† | local | | |
| 166 | Qwen3 30B-A3B Thinking 2507 · 4-bit | 57±1 | local | | |
| 168 | Qwen3.5 Flash | 57±1* | api $0.17/M | | |
| 169 | Olmo 3 7B Think · 8-bit | 57±1† | local | | |
| 171 |  Phi-4-reasoning-plus 14B · 8-bit | 57±1 | local | | |
| 172 | Qwen3-VL 32B Instruct · 4-bit | 57±2*† | local | | |
| 173 | GLM-4.7-Flash · 4-bit | 57±2* | local | | |
| 174 | Gemma 4 12B · 8-bit | 57±2 | local | | |
| 176 |  MiMo-V2.5 (310B-A15B) | 57±2* | api $0.21/M | | |
| 177 | Nemotron Nano 9B v2 · 8-bit | 57±1 | local | | |
| 179 | Qwen3-VL 30B-A3B Instruct · 4-bit | 56±2*† | local | | |
| 180 | MiniMax-M2.7 | 56±2* | api $0.53/M | | |
| 181 | DeepSeek R1 Distill Llama 70B · 2-bit | 56±1 | local | | |
| 184 |  OpenThinker3 7B · 16-bit | 56±1 | local | | |
| 185 | OpenThinker2 32B · 4-bit | 56±2† | local | | |
| 186 | Qwen3 32B · 4-bit | 56±1 | local | | |
| 187 | Qwen3 8B · 8-bit | 55±1 | local | | |
| 188 | DeepSeek R1 Distill Qwen 32B · 4-bit | 55±1 | local | | |
| 189 | DeepSeek R1 Distill Qwen 14B · 8-bit | 55±1 | local | | |
| 190 | Phi-4-reasoning 14B · 8-bit | 55±2† | local | | |
| 191 | GPT-4.1 mini | 55±1 | api $0.70/M | | |
| 192 |  Mistral Medium 3.5 (128B) | 55±2* | api $3.00/M +local | | |
| 194 |  Solar Open 100B (102B-A12B) · 2-bit | 55±2*† | local (slow) | | |
| 196 | Qwen3 4B Instruct 2507 · 8-bit | 54±1 | local | | |
| 197 | Phi-4-mini-reasoning 3.8B · 8-bit | 54±2*† | local | | |
| 198 | Magistral Small 1.2 · 4-bit | 54±1 | local | | |
| 201 | OpenThinker 32B · 4-bit | 53±1† | local | | |
| 202 | Qwen3 4B Thinking 2507 · 8-bit | 53±2 | local | | |
| 203 | GPT-4.1 | 53±1 | api $3.50/M | | |
| 206 |  QED-Nano · 16-bit | 52±2 | local | | |
| 207 | Qwen3-VL 8B Instruct · 8-bit | 52±2 | local | | |
| 210 | DeepSeek R1 Distill Qwen 7B · 8-bit | 52±1 | local | +1 | |
| 211 | EXAONE Deep 32B · 4-bit | 52±2† | local | | |
| 212 | Nemotron 3 Nano 4B · 8-bit | 51±1 | local | | |
| 213 |  s1.1 32B · 4-bit | 51±1 | local | | |
| 214 | Spark-X2.5 1.7B · 16-bit | 51±2† | local | | |
| 217 | Qwen3 30B-A3B (Apr 2025, hybrid) · 4-bit | 50±1 | local | | |
| 218 | Qwen2.5-VL 72B Instruct · 4-bit | 50±4*† | local (slow) | | |
| 219 | EXAONE Deep 7.8B · 4-bit | 50±2† | local | | |
| 220 | OpenThinker2 7B · 4-bit | 50±2† | local | | |
| 221 | Llama 4 Maverick (402B-A17B) | 50±1 | api $0.42/M | | |
| 222 | Gemma 3 27B · 4-bit | 50±1 | local | | |
| 223 | Ministral 3 14B · 8-bit | 50±3*† | local | | |
| 224 | Magistral Small 1.0 (2506) · 8-bit | 50±1 | local | | |
| 225 | Mistral Small 3.2 24B · 8-bit | 50±2* | local | | |
| 228 | Olmo 3 7B Instruct · 8-bit | 49±1† | local | | |
| 230 | Qwen3 235B-A22B (Apr 2025, hybrid) | 49±1 | api $1.23/M | | |
| 231 | Gemma 3 12B · 8-bit | 49±1 | local | | |
| 232 | DeepSeek R1 Distill Llama 8B · 8-bit | 49±1 | local | | |
| 233 | Magistral Small 1.1 (2507) · 8-bit | 49±2*† | local | | |
| 235 | Qwen3-VL 4B Instruct · 8-bit | 49±2*† | local | | |
| 237 | GPT-4.1 nano | 49±1 | api $0.17/M | | |
| 238 | Command A Reasoning (08-2025) · 2-bit | 49±4*† | local (slow) | | |
| 239 |  LFM2.5 1.2B Thinking · 8-bit | 49±3† | local | | |
| 240 | Ministral 3 8B · 8-bit | 48±2 | local | | |
| 242 | LFM2.5 8B-A1B · 8-bit | 48±1 | local | | |
| 243 | Qwen2.5-Math 72B Instruct · 4-bit | 48±2† | local (slow) | | |
| 244 | Qwen3 4B · 8-bit | 48±1 | local | | |
| 245 | LFM2 24B-A2B · 8-bit | 48±2 | local | | |
| 246 | Qwen Plus (2025-01-25) | 48±1* | api $0.60/M | | |
| 247 | Pixtral Large (2411) · 2-bit | 48±7*† | local (slow) | | |
| 248 | DeepSeek V3 (Dec '24) | 48±1 | api $0.46/M | | |
| 249 | Phi-4 14B · 8-bit | 48±1 | local | | |
| 251 | Qwen3-VL 2B Thinking · 8-bit | 48±2*† | local | | |
| 254 | Qwen2.5 72B Instruct · 2-bit | 47±1 | local | | |
| 255 | Llama 4 Scout (109B-A17B) | 47±1* | api $0.15/M +local | | |
| 257 | Qwen2-Math 72B Instruct · 4-bit | 47±2† | local (slow) | | |
| 258 | Granite 4.0 H Small (32B-A9B) · 4-bit | 47±1 | local | | |
| 260 | Gemini 2.5 Flash-Lite | 46±6* | api $0.17/M | | |
| 261 | Tulu 2 70B · 2-bit | 46±5*† | local | | |
| 262 | OpenThinker 7B · 8-bit | 46±2† | local | | |
| 264 | EXAONE Deep 2.4B · 4-bit | 46±2† | local | | |
| 265 | Qwen2.5-Math 7B Instruct · 4-bit | 46±2† | local | | |
| 266 | Qwen2.5 32B Instruct · 4-bit | 46±1 | local | | |
| 267 | Llama 3.1 Nemotron 51B Instruct · 4-bit | 46±5*† | local (slow) | | |
| 268 | LFM2.5 2.6B · 8-bit | 46±1 | local | | |
| 271 | Tulu 2 DPO 70B · 4-bit | 45±5*† | local (slow) | | |
| 272 | GPT-4o (2024-08-06) | 45±1 | api $4.38/M | | |
| 273 | Qwen2.5 14B Instruct · 8-bit | 45±2 | local | | |
| 274 | Granite 4.0 H Tiny (7B-A1B) · 4-bit | 45±2† | local | | |
| 275 | DeepSeek R1 Distill Qwen 1.5B · 8-bit | 45±1 | local | | |
| 278 | GPT-4o (2024-05-13) | 45±1 | api $7.50/M | | |
| 279 | GPT-4o mini (2024-07-18) | 45±1 | api $0.26/M | +2 | |
| 281 | Llama 3.1 Tulu 3 70B · 4-bit | 45±3*† | local (slow) | | |
| 282 | Qwen2.5-VL 7B Instruct · 8-bit | 45±3*† | local | | |
| 283 | Command A (03-2025) | 44±3*† | api $4.38/M +local | | |
| 284 | Gemma 4 E4B · 8-bit | 44±1 | local | | |
| 286 | Qwen2-VL 72B Instruct · 4-bit | 44±3 | local (slow) | | |
| 287 | Llama 3.1 405B Instruct | 44±1 | api $2.40/M | | |
| 288 | GPT-4o (2024-11-20) | 44±1 | api $4.38/M | | |
| 290 | Mixtral 8x7B v0.1 (base) · 4-bit | 44±2*† | local | | |
| 291 | EXAONE 4.0 1.2B · 8-bit | 44±2*† | local | | |
| 292 | Qwen2.5 7B Instruct · 8-bit | 44±1 | local | | |
| 293 |  SmolLM3 3B · 8-bit | 44±2 | local | | |
| 294 | Gemma 3 4B · 8-bit | 43±2 | local | | |
| 295 | Mistral Small 3.1 24B (2503) · 8-bit | 43±1 | local | | |
| 296 | GPT-4 Turbo (2024-04-09) | 43±1 | api $15.00/M | | |
| 297 | Granite 3.3 8B Instruct · 8-bit | 43±1 | local | | |
| 299 | Mistral Large 2 (2407) · 2-bit | 43±1 | local (slow) | | |
| 302 |  InternVL2-4B · 16-bit | 43±5* | local | | |
| 304 |  Xwin-LM 70B v0.1 · 2-bit | 43±5* | local | | |
| 305 | Qwen3 1.7B · 8-bit | 43±1 | local | +1 | |
| 306 | Mistral Small 3 24B (2501) · 8-bit | 43±1* | local | | |
| 307 | Granite 4.0 H Micro 3B · 4-bit | 43±2† | local | | |
| 308 | Llama 3.1 Tulu 3 70B DPO · 2-bit | 43±1 | local | | |
| 309 | Llama 3.3 70B Instruct · 2-bit | 42±1 | local | | |
| 310 | Ministral 3 3B · 8-bit | 42±1 | local | | |
| 312 | Qwen2-Math 7B Instruct · 8-bit | 42±2† | local | | |
| 313 | Granite 4.0 Micro 3B · 8-bit | 42±2 | local | | |
| 315 | Granite 8B Code Base (4K) · 4-bit | 42±2*† | local | | |
| 318 | Qwen2 72B Instruct · 4-bit | 41±1 | local (slow) | | |
| 319 | Llama 3.1 70B Instruct · 2-bit | 41±1 | local | | |
| 322 | Gemma 4 E2B · 8-bit | 41±1 | local | | |
| 323 | Qwen2.5-Math 1.5B Instruct · 4-bit | 41±2† | local | | |
| 325 | Phi-4-mini 3.8B · 8-bit | 41±2† | local | | |
| 327 | Phi-2 (2.7B) · 4-bit | 40±3* | local | | |
| 328 | Molmo 2 8B · 8-bit | 40±7*† | local | | |
| 329 | Pixtral 12B (2409) · 16-bit | 40±3*† | local | | |
| 330 |  StarCoder2 15B · 8-bit | 40±2*† | local | | |
| 331 | LFM2 2.6B · 4-bit | 40±3*† | local | | |
| 332 | Qwen3.5 2B · 8-bit | 40±1 | local | | |
| 333 | Llama 3.1 Nemotron 70B Instruct · 4-bit | 40±3* | local (slow) | | |
| 334 | Qwen2.5-VL 3B Instruct · 8-bit | 40±3*† | local | | |
| 336 | Qwen3-VL 2B Instruct · 8-bit | 40±2*† | local | | |
| 337 | Gemma 7B (base) · 8-bit | 40±2 | local | | |
| 338 |  InternLM2.5 7B Chat · 8-bit | 39±3*† | local | | |
| 339 | Qwen2-Math 1.5B Instruct · 8-bit | 39±2† | local | | |
| 340 | CodeGemma 7B (v1.0) · 8-bit | 39±2* | local | | |
| 342 | LFM2 1.2B · 4-bit | 39±3*† | local | | |
| 343 | WizardLM 70B V1.0 · 4-bit | 38±4* | local (slow) | | |
| 345 | Llama 3.2 11B Vision Instruct · 8-bit | 38±3* | local | | |
| 347 | LFM2.5 1.2B Instruct · 4-bit | 38±2† | local | | |
| 348 | Gemma 2 27B IT · 4-bit | 38±1 | local | | |
| 350 | Llama 2 70B (base) · 4-bit | 38±2 | local (slow) | | |
| 351 | Phi-3-vision-128k-instruct · 16-bit | 38±6*† | local | | |
| 352 | Mathstral 7B v0.1 · 8-bit | 38±3*† | local | | |
| 353 | Granite 3.3 2B Instruct · 8-bit | 38±2† | local | | |
| 355 | Phi-3.5-vision-instruct · 16-bit | 38±7*† | local | | |
| 356 | DeepSeek-Coder 33B Base · 4-bit | 38±2* | local | | |
| 357 | GPT-4 (0613) | 37±1 | api $37.50/M | | |
| 358 | Tulu 2 DPO 13B · 16-bit | 37±5*† | local | | |
| 359 | Llama 3 70B Instruct · 2-bit | 37±1 | local | | |
| 360 |  Jamba 1.5 Mini · 4-bit | 37±4*† | local (slow) | | |
| 361 |  Yi 1.5 34B Chat · 4-bit | 37±1 | local | | |
| 362 | Llama 3.1 Tulu 3 8B · 8-bit | 37±1 | local | | |
| 363 |  Hermes 2 Theta Llama-3 70B · 4-bit | 37±1* | local (slow) | | |
| 364 | Granite 3.2 8B Instruct · 8-bit | 37±2† | local | | |
| 366 | Granite 4.0 1B · 16-bit | 37±2 | local | | |
| 367 | Qwen2.5 3B Instruct · 8-bit | 37±1 | local | | |
| 368 | OLMo 2 32B Instruct (0325) · 4-bit | 37±1† | local | | |
| 369 | Llama 3.1 8B Instruct · 8-bit | 37±1 | local | | |
| 370 | Granite 3.1 8B Instruct · 8-bit | 36±2† | local | | |
| 371 | OLMo 2 13B Instruct (1124) · 8-bit | 36±2*† | local | | |
| 372 | Gemma 2 9B IT · 8-bit | 36±1 | local | | |
| 373 | Tulu 2 13B · 16-bit | 36±4*† | local | | |
| 374 | Granite 3.0 8B Instruct · 8-bit | 36±2*† | local | | |
| 377 | Granite 4.0 H 1B · 8-bit | 36±2 | local | | |
| 378 |  Athene-70B · 2-bit | 35±1* | local | | |
| 379 | Qwen3 0.6B · 8-bit | 35±2 | local | | |
| 380 | Phi-3 Medium 128K Instruct · 8-bit | 35±1* | local | | |
| 381 | Code Llama 34B (base) · 4-bit | 35±2* | local | | |
| 382 | Qwen2 7B Instruct · 8-bit | 35±1 | local | | |
| 384 | Ministral 8B (2410) · 16-bit | 34±1 | local | | |
| 385 | Code Tulu 2 13B · 16-bit | 34±5*† | local | | |
| 387 | LFM2 700M · 4-bit | 34±3*† | local | | |
| 390 | Phi-3.5-mini-instruct · 8-bit | 34±1 | local | | |
| 392 | Qwen2-VL 7B Instruct · 8-bit | 34±2 | local | | |
| 393 | Code Llama 34B Instruct · 4-bit | 33±5* | local | | |
| 394 | Yi 1.5 9B Chat · 8-bit | 33±1* | local | | |
| 396 | Granite 3.0 2B Instruct · 8-bit | 33±2*† | local | | |
| 397 | OLMo 2 7B Instruct (1124) · 8-bit | 33±2† | local | | |
| 398 | Qwen3.5 0.8B · 8-bit | 33±4† | local | | |
| 399 | Qwen2.5 1.5B Instruct · 8-bit | 33±2 | local | | |
| 400 | GPT-3.5 Turbo (0125) | 33±1 | api $0.75/M | | |
| 401 | Mistral 7B v0.1 (base) · 16-bit | 33±1 | local | | |
| 402 | Llama 3.2 3B Instruct · 8-bit | 32±1 | local | | |
| 404 | Mistral NeMo Instruct (2407) · 8-bit | 32±1* | local | | |
| 406 | Tulu 2 DPO 7B · 16-bit | 32±5*† | local | | |
| 408 | Tulu 2 7B · 16-bit | 32±5*† | local | | |
| 409 | Phi-3-mini-4k-instruct (June 2024 update) · 8-bit | 32±1* | local | | |
| 410 | Command R Plus (04-2024) · 2-bit | 32±1* | local (slow) | | |
| 411 | Code Tulu 2 7B · 16-bit | 32±5*† | local | | |
| 412 | Gemma 3 1B · 4-bit | 31±1 | local | | |
| 413 | DeepSeek-Coder 6.7B Base · 8-bit | 31±2* | local | | |
| 414 | Granite 3.2 2B Instruct · 8-bit | 31±2† | local | | |
| 415 | Granite 3.1 2B Instruct · 8-bit | 31±2† | local | | |
| 417 |  Stable Code 3B · 8-bit | 31±2* | local | | |
| 418 | Mixtral 8x7B Instruct v0.1 · 4-bit | 30±1* | local | | |
| 419 | StarCoder2 7B · 8-bit | 30±2 | local | | |
| 420 | Granite 3.1 3B-A800M Instruct · 8-bit | 30±4*† | local | | |
| 421 | Zephyr 7B beta · 8-bit | 30±5* | local | | |
| 422 | Granite 3.0 3B-A800M Instruct · 8-bit | 30±2*† | local | | |
| 423 | OLMo 2 1B Instruct (0425) · 16-bit | 29±2*† | local | | |
| 425 | Code Llama 13B (base) · 8-bit | 29±2* | local | | |
| 426 | Llama 3 8B Instruct · 8-bit | 29±1 | local | | |
| 427 | WizardLM 13B V1.2 · 4-bit | 29±4* | local | | |
| 428 | DeepSeek LLM 67B Chat · 2-bit | 29±1* | local | | |
| 429 |  Apertus 8B Instruct (2509) · 16-bit | 29±3* | local | | |
| 430 | Code Llama 13B Instruct · 8-bit | 28±5* | local | | |
| 431 | SmolLM2 1.7B Instruct · 8-bit | 27±2 | local | | |
| 432 | Yi 34B Chat · 4-bit | 27±1* | local | | |
| 433 | LFM2 350M · 4-bit | 27±3*† | local | | |
| 434 | Falcon 40B (base) · 4-bit | 27±3* | local | | |
| 435 | Gemma 2B (base) · 8-bit | 27±4*† | local | | |
| 436 | Llama 2 70B Chat · 4-bit | 26±1 | local (slow) | | |
| 437 | Llama 2 13B (base) · 8-bit | 26±2 | local | | |
| 438 | StarCoder2 3B · 8-bit | 26±2*† | local | | |
| 439 | Llama 3.2 1B Instruct · 4-bit | 25±1 | local | | |
| 440 | Code Llama 7B Instruct · 8-bit | 25±5* | local | | |
| 441 |  Llama-3-Instruct-8B-SimPO-v0.2 · 16-bit | 25±2* | local | | |
| 442 | Granite 3.1 1B-A400M Instruct · 8-bit | 25±4*† | local | | |
| 443 | Mistral 7B Instruct v0.3 · 8-bit | 25±1 | local | | |
| 444 | Qwen2 0.5B Instruct · 8-bit | 24±5*† | local | | |
| 445 | Molmo 7B-D (0924) · 4-bit | 24±3 | local (slow) | | |
| 446 | Gemma 2 2B IT · 8-bit | 24±1 | local | | |
| 447 | Command R (03-2024) · 4-bit | 23±1* | local | | |
| 448 | Llama 2 7B Chat · 8-bit | 23±5* | local | | |
| 449 | Qwen2-VL 2B Instruct · 4-bit | 23±3*† | local | | |
| 450 | StarCoderBase 15.5B · 8-bit | 23±2*† | local | | |
| 451 | Qwen2 1.5B Instruct · 8-bit | 23±1 | local | | |
| 452 | Granite 3.0 1B-A400M Instruct · 8-bit | 23±3*† | local | | |
| 453 | Code Llama 7B (base) · 8-bit | 22±2 | local | | |
| 455 | Granite 4.0 H 350M · 8-bit | 20±2† | local | | |
| 456 | Granite 4.0 350M · 16-bit | 19±2 | local | | |
| 457 | Llama 2 13B Chat · 8-bit | 19±3* | local | | |
| 458 | Llama 2 7B (base) · 8-bit | 18±2 | local | | |
| 460 | DeepSeek-Coder 1.3B Base · 8-bit | 17±3* | local | | |
| 461 | StarCoderBase 7B · 4-bit | 17±2 | local | | |
| 462 | StarCoderBase 3B · 4-bit | 13±3*† | local | | |
| — | DeepSeek LLM 7B Chat · q4_0 | not yet calibrated | local | | |
| — | DeepSeek-V2 Lite Chat (16B-A2.4B) · q8_0 | not yet calibrated | local | | |
| — |  Eurus-2-7B-PRIME · Q8_0 | not yet calibrated | local | | |
| — | Eurus-2-7B-SFT · Q8_0 | not yet calibrated | local | | |
| — |  Flan-T5 Base · fp32 | not yet calibrated | local | | |
| — | Flan-T5 Large · fp32 | not yet calibrated | local | | |
| — | Flan-T5 Small · fp32 | not yet calibrated | local | | |
| — | Flan-T5 XL · fp32 | not yet calibrated | local | | |
| — | Flan-T5 XXL · fp32 | not yet calibrated | local (slow) | | |
| — | Gemma 3n E2B IT · q8_0 | not yet calibrated | local | | |
| — | Gemma 3n E4B · q8_0 | not yet calibrated | local | | |
| — | GPT-2 XL (1.5B) · fp32 | not yet calibrated | local | | |
| — |  GPT-Neo 1.3B · fp32 | not yet calibrated | local | | |
| — | GPT-Neo 2.7B · fp32 | not yet calibrated | local | | |
| — | Granite 4.1 30B · q4_k_m | not yet calibrated | local | | |
| — | Granite 4.1 3B · q4_k_m | not yet calibrated | local | | |
| — | Granite 4.1 8B · q8_0 | not yet calibrated | local | | |
| — |  Hunyuan Hy3 (295B-A21B) | not yet calibrated | api $0.23/M | | |
| — | Hy4 preview (770B-A49B) | not yet calibrated | api $1.25/M | | |
| — |  K2 Horizon 7B · gguf-q8_0 | not scored yet | local | ||
| — |  Ling 3.0 Flash (124B-A5B) | not yet calibrated | api $0.090/M | | |
| — | Llama 3.3 Nemotron Super 49B v1 · ud-q2_k_xl | not yet calibrated | local | | |
| — | MiniMax-M1 (80k) | not yet calibrated | api $0.96/M | | |
| — | Ministral 3 14B Reasoning 2512 · q4_k_m | not yet calibrated | local | | |
| — | Ministral 3 3B Reasoning 2512 · q4_k_m | not yet calibrated | local | | |
| — | Ministral 3 8B Reasoning 2512 · q4_k_m | not yet calibrated | local | | |
| — | Mistral Small 4 (119B-A6.5B) | not scored yet | api $0.26/M +local | ||
| — | Muse Glimmer 30B · q4_k_m | not scored yet | local | ||
| — |  Nova Premier | not yet calibrated | api $5.00/M |  | |
| — | Nova Pro | not yet calibrated | api $1.40/M | | |
| — | OLMo 7B · fp32 | not yet calibrated | local | | |
| — | OLMo 7B Instruct · bf16 | not yet calibrated | local | | |
| — |  OpenChat 3.5 · q8_0 | not yet calibrated | local | | |
| — | OpenHermes 2.5 Mistral 7B · q8_0 | not yet calibrated | local | | |
| — | Qwen2.5 0.5B Instruct · q8_0 | not yet calibrated | local | | |
| — | SmolLM2 360M Instruct · fp16 | not yet calibrated | local | | |
| — |  Trinity-Large-Thinking | not yet calibrated | api $0.39/M | |
+ 103 more don't fit your machine or budget · show them
+ 2 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them
509 of 511 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