Best AI models for chat and writing for a 32GB graphics card
A leaderboard of the best AI models for chat and writing for a 32GB graphics card (with 32GB of system RAM). The best one that runs on it right now is Gemma 4 31B, followed by Nemotron Cascade 2 30B-A3B and Qwen3.8 27B. Models are ranked on human votes from LMArena and on writing and instruction-following tests. Models that run on it show the download that fits and the command to run it.
370 models · scores combine 19 benchmarks · updated how it works · find a leaderboard · benchmark your machine
Scores use reviewed chat & writing 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 |  Gemini 4 Argon (limited preview) | 78±1* | limited preview |  | |
| 2 |  Claude Fable 5 | 76±1* | api $20.00/M | | |
| 3 | Claude Opus 4.6 | 76±1* | api $10.00/M |  | |
| 4 | Claude Opus 5.5 | 76±1* | api $8.00/M | | |
| 5 | Claude Fable 5.1 | 75±1* | api $20.00/M | | |
| 6 | Claude Opus 4.7 | 75±1* | api $10.00/M | | |
| 7 | Gemini 3.8 Flash | 75±1* | api $1.50/M | | |
| 8 |  Muse Spark 1.3 | 75±1* | api $2.00/M | | |
| 9 | Muse Spark 1.2 | 74±1* | api $2.00/M | | |
| 10 | Muse Spark 1.1 | 74±1* | api $2.00/M | | |
| 11 | Claude Opus 5 | 74±1* | api $10.00/M | | |
| 13 |  Kimi K3 | 74±1* | api $6.00/M | | |
| 14 | Gemini 3.7 Flash | 74±1* | api $1.50/M | | |
| 15 | Gemini 3.1 Pro (Preview) | 74±1* | api $4.50/M | | |
| 17 |  GPT-5.6 Sol | 73±1* | api $8.00/M | | |
| 18 | Gemini 3.6 Flash | 73±1* | api $1.50/M | | |
| 19 | GPT-6.1 Sol | 73±1* | api $4.00/M | | |
| 20 |  Qwen3.8 Max (2.4T-A95B) | 73±1* | api $3.00/M | | |
| 21 | GPT-5.5 | 73±1* | api $11.25/M | | |
| 22 |  MiMo-V2.6-Pro | 73±1* | api $0.54/M | | |
| 23 | Claude Opus 4.8 | 73±1* | api $10.00/M | | |
| 24 |  GLM-5.3 | 73±1* | api $2.15/M | | |
| 25 | GPT-6 Astra | 72±1* | api $20.00/M | | |
| 26 | Gemini 3.5 Flash | 72±1* | api $3.38/M | | |
| 27 | GLM-5.2 | 72±1* | api $2.15/M | | |
| 28 | GPT-5.4 | 72±1* | api $5.63/M | | |
| 29 |  DeepSeek V4.1 Flash | 72±1* | api $0.26/M | | |
| 30 | GLM-5.3-Flash | 72±1* | api $0.24/M | | |
| 31 | Claude Opus 4.5 | 72±1* | api $10.00/M | | |
| 33 | Claude Sonnet 4.6 | 72±1* | api $6.00/M | | |
| 34 | Gemini 3 Flash (Preview) | 72±1* | api $1.13/M | | |
| 35 |  Grok 4.20 (0309 reasoning) | 72±1* | api $1.56/M | | |
| 36 | Grok 4.20 Multi-Agent (0309) | 72±1* | api $1.56/M | | |
| 37 | Claude Sonnet 5.5 | 72±1* | api $4.00/M | | |
| 38 | MiMo-V2.5-Pro (1T-A42B) | 71±1* | api $0.65/M | | |
| 39 | GPT-5.6 Terra | 71±1* | api $4.50/M | | |
| 40 | Grok 4.5 | 71±1* | api $3.00/M | | |
| 41 | DeepSeek V4 Pro (0813) | 71±1* | api $1.63/M | | |
| 42 | GLM-5.1 | 71±1* | api $2.15/M | | |
| 43 | Claude Sonnet 5 | 70±1* | api $4.00/M | | |
| 44 | Kimi K2.6 | 70±1* | api $1.71/M | | |
| 45 | Qwen3.6 Max Preview | 70±1* | api $2.92/M | | |
| 46 | GPT-6 Sol | 70±1* | api $4.00/M | | |
| 47 | GLM-5 | 70±1* | api $1.55/M | | |
| 48 | Claude Sonnet 4.5 | 70±1* | api $6.00/M | | |
| 50 | GPT-5.1 | 70±1* | api $3.44/M | | |
| 51 | Qwen3.7 Plus | 70±1* | api $0.70/M | | |
| 52 |  Hunyuan Hy3 (295B-A21B) | 70±1* | api $0.23/M | | |
| 53 | Gemini 3.5 Flash-Lite | 70±1* | api $0.85/M | | |
| 55 | Grok 4.6 | 69±1* | api $3.00/M | | |
| 57 |  Gemma 4 31B · 4-bit | 69±1* | local |  | |
| 58 | GPT-5.6 Luna | 69±1* | api $0.45/M | | |
| 59 | MiMo-V2.6-Flash (309B-A15B) | 69±1* | api $0.17/M | | |
| 60 | Kimi K2.5 | 69±1* | api $1.20/M | | |
| 63 | GPT-5.4 mini | 69±1* | api $1.69/M | | |
| 64 | Gemini 2.5 Pro | 68±1* | api $3.44/M | | |
| 65 | Qwen3.6 Plus | 68±1* | api $1.13/M | | |
| 66 | GPT-6 Luna | 68±1* | api $0.20/M | | |
| 67 | Grok 4.7 | 68±1* | api $3.00/M | | |
| 69 | Grok 4.3 | 68±1* | api $1.56/M | | |
| 70 | GLM-4.7 | 68±1* | api $1.00/M | | |
| 71 |  Inkling (975B-A41B) | 68±1* | api $1.73/M | | |
| 72 |  Nemotron Cascade 2 30B-A3B · 4-bit | 68±4*† | local | | |
| 73 | Qwen3.5 397B-A17B | 68±1* | api $1.09/M | | |
| 74 |  MiniMax-M3 | 68±1* | api $0.48/M | | |
| 77 | Qwen3.8 27B · 4-bit | 67±1* | local | | |
| 78 | Gemma 4 26B-A4B · 8-bit | 67±1* | local | | |
| 79 | GPT-5.2 | 67±1* | api $4.81/M | | |
| 80 | o3 | 67±1 | api $3.50/M |  | |
| 81 | GPT-5 | 67±1* | api $3.44/M | | |
| 82 | MiMo-V2.5 (310B-A15B) | 67±1* | api $0.21/M | | |
| 83 | Gemma 4 12B · 8-bit | 67±6* | local | | |
| 84 | GLM-5V-Turbo | 67±1* | api $1.90/M | | |
| 85 | Gemini 3.1 Flash-Lite | 67±1* | api $0.56/M | | |
| 87 | Kimi K2 Thinking | 66±1* | api $1.07/M | | |
| 88 | Grok 4.1 Fast (Reasoning) | 66±1* | api $0.28/M | | |
| 89 |  Mistral Large 4 (public preview) | 66±1* | api $2.06/M | | |
| 90 | Mistral Medium 3.5 (128B) | 66±1* | api $3.00/M +local | | |
| 91 | Nemotron 3 Ultra 550B-A55B | 66±1* | api $1.35/M | | |
| 93 | Muse Glimmer 30B · 4-bit | 66±1* | local | | |
| 95 | GLM-4.6 | 66±1* | api $1.00/M | | |
| 96 | DeepSeek V3.2 | 65±1* | api $0.29/M | | |
| 97 | Qwen3-Max (2025-09-23) | 65±1* | api $2.40/M | | |
| 98 | DeepSeek R1 0528 | 65±1* | api $0.91/M | | |
| 99 | Qwen3 235B-A22B Instruct 2507 | 65±1 | api $0.20/M | | |
| 101 | Kimi K2 Instruct (0711) | 65±1* | api $1.00/M | | |
| 103 | Qwen3.5 122B-A10B | 65±1* | api $1.10/M | | |
| 104 | MiniMax-M2.7 | 64±1* | api $0.53/M | | |
| 105 | Claude Haiku 4.5 | 64±1* | api $2.00/M | | |
| 107 | DeepSeek V3.1 | 64±1 | api $0.42/M | | |
| 108 | Mistral Large 3 (675B-A41B) | 64±1* | api $0.75/M | | |
| 111 |  Spark-X2.5 4B · 16-bit | 64±6*† | local | | |
| 112 | GPT-4.1 | 64±1 | api $3.50/M | | |
| 113 | Qwen3.5 27B · 4-bit | 64±1* | local | | |
| 114 | GLM-4.5 | 64±1 | api $1.00/M | | |
| 117 | Gemini 2.5 Flash | 63±1* | api $0.85/M | | |
| 119 | Inkling Small (266B-A12B) | 63±1* | api $0.64/M | | |
| 121 | o1 (2024-12-17) | 63±1 | api $26.25/M | | |
| 124 | Nemotron-Cascade-8B · 8-bit | 63±7*† | local | | |
| 125 | GPT-5.4 nano | 63±1* | api $0.46/M | | |
| 126 | Qwen3 235B-A22B Thinking 2507 | 63±1 | api $0.97/M | | |
| 128 | Qwen3-Next 80B-A3B Instruct · 2-bit | 63±1 | local | | |
| 129 | DeepSeek R1 (Jan '25) | 63±1 | api $4.00/M | | |
| 130 | Qwen3.5 Flash | 62±1* | api $0.17/M | | |
| 131 |  LFM2.5 2.6B · 8-bit | 62±7* | local | | |
| 132 | o4-mini | 62±1 | api $1.93/M | | |
| 134 |  Step 3.5 Flash (196B-A11B) | 62±1* | api $0.15/M | | |
| 136 | Qwen3.5 35B-A3B · 4-bit | 62±1 | local | | |
| 137 | MiniMax-M2.5 | 61±1* | api $0.53/M | | |
| 139 | GPT-5 mini | 61±1* | api $0.69/M | | |
| 141 | Qwen3-Coder 480B-A35B Instruct | 61±1* | api $0.67/M | | |
| 143 |  Solar Pro 4 | 61±1* | api $0.53/M | | |
| 146 | LFM2.5 8B-A1B · 8-bit | 60±5* | local | | |
| 148 | Qwen3 30B-A3B Instruct 2507 · 4-bit | 60±1 | local | | |
| 149 | GPT-4.1 mini | 60±1 | api $0.70/M | | |
| 150 | GLM-4.6V | 60±1* | api $0.45/M +local | | |
| 152 |  LongCat 2.0 (1.6T-A48B) | 60±7*† | api $1.30/M | | |
| 154 | Spark-X2.5 1.7B · 16-bit | 60±5*† | local | | |
| 156 | Qwen3 235B-A22B (Apr 2025, hybrid) | 59±1 | api $1.23/M | | |
| 157 | GLM-4.5-Air | 59±1* | api $0.42/M +local | | |
| 159 |  Gemma 2 9B IT DPO (Princeton) · 16-bit | 59±2* | local | | |
| 160 | Qwen3-Next 80B-A3B Thinking · 2-bit | 59±1 | local | | |
| 161 | o3-mini | 59±1 | api $1.93/M | | |
| 163 |  EXAONE 4.5 33B · 4-bit | 59±4*† | local | | |
| 164 | LFM2.5 1.2B Thinking · 8-bit | 59±6*† | local | | |
| 165 |  Trinity-Large-Thinking | 58±1* | api $0.39/M | | |
| 166 | GLM-4.7-Flash · 4-bit | 58±1* | local | | |
| 167 | MiniMax-M1 (80k) | 58±1* | api $0.96/M | | |
| 169 | Qwen3.5 9B · 8-bit | 58±3 | local | | |
| 170 | Gemma 3 27B · 4-bit | 58±1 | local | | |
| 171 | Qwen3 30B-A3B Thinking 2507 · 4-bit | 58±2 | local | | |
| 172 | Nemotron 3 Super 120B-A12B | 58±1* | api $0.28/M | | |
| 173 | DeepSeek V3 (Dec '24) | 57±1 | api $0.46/M | | |
| 175 |  Granite 4.1 30B · 4-bit | 57±5*† | local | | |
| 176 |  Phi-4-reasoning-plus 14B · 8-bit | 57±3† | local | | |
| 177 | Qwen3-VL 32B Thinking · 4-bit | 57±3*† | local |  | |
| 178 | Mistral Small 3.2 24B · 8-bit | 57±1* | local | | |
| 179 |  INTELLECT-3 · 2-bit | 57±1* | local (slow) | | |
| 180 |  Command A (03-2025) | 57±1* | api $4.38/M +local | | |
| 182 | GLM-4.5V | 57±1* | api $0.90/M | | |
| 183 | gpt-oss-120b | 56±1* | api $0.26/M | | |
| 186 | Nemotron 3.5 Lightning 30B-A3B · 4-bit | 56±1* | local | | |
| 187 | Qwen3.5 4B · 8-bit | 56±3 | local | | |
| 188 | Nemotron 3 Nano 4B · 8-bit | 56±6* | local | | |
| 189 | LFM2.5 1.2B Instruct · 4-bit | 56±6*† | local | | |
| 190 | Qwen3 32B · 4-bit | 56±1 | local | | |
| 191 | Qwen Plus (2025-01-25) | 56±1* | api $0.60/M | | |
| 192 | QwQ-32B · 4-bit | 56±1 | local | | |
| 193 |  Mellum2.1 12B-A2.5B Thinking · 16-bit | 56±8*† | local | | |
| 194 | MiniMax-M2 | 55±1* | api $0.53/M | | |
| 195 | Llama 3.3 Nemotron Super 49B v1.5 · 2-bit | 55±1* | local | | |
| 198 | Gemma 3 12B · 8-bit | 55±1* | local | | |
| 199 | Granite 4.2 30B · 4-bit | 55±1* | local | | |
| 200 | GPT-4o (2024-05-13) | 55±1 | api $7.50/M | | |
| 201 | Qwen3-VL 30B-A3B Instruct · 4-bit | 55±3*† | local | | |
| 202 |  MiniCPM5 2B · 8-bit | 55±6*† | local | | |
| 203 | GPT-5 nano | 55±1* | api $0.14/M | | |
| 205 | Phi-4-reasoning 14B · 8-bit | 54±3† | local | | |
| 206 |  Nova 2 Lite | 54±1* | api $0.94/M | | |
| 207 | GPT-4o (2024-08-06) | 54±1 | api $4.38/M | | |
| 208 | Qwen3 8B · 8-bit | 54±3*† | local | | |
| 209 | Qwen3 4B Instruct 2507 · 8-bit | 54±2† | local | | |
| 210 | Qwen3-VL 32B Instruct · 4-bit | 54±3*† | local | | |
| 211 | Llama 3.3 Nemotron Super 49B v1 · 2-bit | 54±1 | local | | |
| 212 | Nemotron Nano 9B v2 · 8-bit | 54±3 | local | | |
| 213 |  Yi-Lightning | 54±1* | api price unknown | | |
| 214 |  Olmo 3.1 32B Instruct · 4-bit | 54±1* | local | | |
| 215 | Qwen3 4B Thinking 2507 · 8-bit | 53±3*† | local | | |
| 216 | Llama 3.1 405B Instruct | 53±1 | api $2.40/M | | |
| 217 | GPT-4 Turbo (2024-04-09) | 53±1 | api $15.00/M | | |
| 219 | Nemotron-Cascade-14B-Thinking · 8-bit | 53±6*† | local | | |
| 220 | Qwen3-VL 8B Instruct · 8-bit | 53±3*† | local | | |
| 222 | Qwen3 30B-A3B (Apr 2025, hybrid) · 4-bit | 53±1 | local | | |
| 223 | Llama 4 Maverick (402B-A17B) | 53±1 | api $0.42/M | | |
| 224 |  Seed-OSS 36B Instruct · 4-bit | 53±5*† | local | | |
| 226 | Llama 4 Scout (109B-A17B) | 53±1* | api $0.15/M +local | | |
| 227 | EXAONE 4.0 32B · 4-bit | 53±4*† | local | | |
| 228 | Molmo 2 8B · 8-bit | 53±1* | local | | |
| 229 | Qwen3-VL 8B Thinking · 8-bit | 53±3*† | local | | |
| 230 | Olmo 3 7B Think · 8-bit | 53±3 | local | | |
| 231 | gpt-oss-20b · MXFP4 | 52±1* | local | | |
| 232 | GPT-4.1 nano | 52±1 | api $0.17/M | | |
| 233 |  Athene-V2-Chat 72B · 4-bit | 52±1 | local (slow) | | |
| 234 | Qwen3-VL 4B Thinking · 8-bit | 52±3*† | local | | |
| 235 | Gemma 3n E4B · 8-bit | 52±1* | local | | |
| 236 | GPT-4o mini (2024-07-18) | 52±1 | api $0.26/M | | |
| 238 | Qwen3-VL 4B Instruct · 8-bit | 52±3*† | local | | |
| 239 | Mistral Large 2 (2407) · 2-bit | 52±1 | local (slow) | | |
| 240 | Granite 4.0 H Micro 3B · 4-bit | 52±5*† | local | | |
| 242 | Llama 3.3 70B Instruct · 2-bit | 52±1 | local | | |
| 243 | Nemotron 3 Nano 30B-A3B · 4-bit | 52±1* | local | | |
| 244 | Qwen3-VL 30B-A3B Thinking · 4-bit | 51±3*† | local | | |
| 245 | Athene-70B · 2-bit | 51±1* | local | | |
| 246 | Olmo 3 32B Think · 4-bit | 51±1 | local | | |
| 250 | Qwen2.5 72B Instruct · 2-bit | 51±1 | local | | |
| 251 | Llama 3.1 Nemotron 70B Instruct · 4-bit | 51±1 | local (slow) | | |
| 253 | Granite 4.1 8B · 8-bit | 51±1* | local | | |
| 254 | Gemma 2 9B IT SimPO · 16-bit | 51±1* | local | | |
| 256 | Olmo 3 7B Instruct · 8-bit | 50±3 | local | | |
| 258 | Gemma 3 4B · 8-bit | 50±1 | local | | |
| 259 | Granite 4.1 3B · 4-bit | 50±5*† | local | | |
| 260 | Mistral Small 3.1 24B (2503) · 8-bit | 50±1* | local | | |
| 261 | Granite 4.0 Micro 3B · 8-bit | 50±5*† | local | | |
| 262 | Command R7B (12-2024) · 8-bit | 49±3* | local | | |
| 263 |  Flan-T5 XXL · 4-bit | 49±15* | local (slow) | | |
| 265 | Qwen2.5 32B Instruct · 4-bit | 49±2 | local | | |
| 266 | Llama-3-Instruct-8B-SimPO-ExPO · 16-bit | 49±2* | local | | |
| 267 | Granite 4.2 3B · 8-bit | 49±1 | local | | |
| 268 | Granite 4.2 8B · 8-bit | 49±1* | local | | |
| 269 | Olmo 3.1 32B Think · 4-bit | 49±1 | local | | |
| 270 | Llama 3.1 Tulu 3 70B DPO · 2-bit | 49±4*† | local | | |
| 271 | Granite 4.0 H Small (32B-A9B) · 4-bit | 49±1 | local | | |
| 272 | Nova Pro | 49±1* | api $1.40/M | | |
| 274 | Llama-3-Instruct-8B-SimPO · 16-bit | 49±2* | local | | |
| 275 | Granite 4.0 H Tiny (7B-A1B) · 4-bit | 49±4*† | local | | |
| 276 | Gemma 2 27B IT · 4-bit | 48±1 | local | | |
| 277 | Granite 4.0 H 1B · 8-bit | 48±4 | local | | |
| 278 | Llama 3.1 70B Instruct · 2-bit | 48±1 | local | | |
| 280 | Llama 3.1 Tulu 3 70B · 4-bit | 48±1 | local (slow) | | |
| 281 | Llama 3.1 Nemotron 51B Instruct · 4-bit | 48±1* | local (slow) | | |
| 282 |  s1.1 32B · 4-bit | 48±4* | local | | |
| 283 | Mellum2 12B-A2.5B Instruct · 16-bit | 48±4*† | local | | |
| 285 | LFM2 2.6B · 4-bit | 48±6*† | local | | |
| 286 | Qwen3 4B · 8-bit | 47±2 | local | | |
| 288 | Granite 4.0 1B · 16-bit | 47±3 | local | | |
| 289 | OLMo 2 13B Instruct (1124) · 8-bit | 47±3*† | local | | |
| 290 | Qwen3.5 2B · 8-bit | 47±3 | local | | |
| 291 | Qwen2.5 14B Instruct · 8-bit | 47±3 | local | | |
| 295 | Command R Plus (08-2024) | 47±1* | api $4.38/M | | |
| 296 | Mellum2 12B-A2.5B Thinking · 16-bit | 47±3*† | local | | |
| 298 | Mistral Small 3 24B (2501) · 8-bit | 47±1* | local | | |
| 299 | Qwen3-VL 2B Thinking · 8-bit | 47±3*† | local | | |
| 302 | Qwen2.5 Coder 32B Instruct · 4-bit | 46±1* | local | | |
| 303 | Aya Expanse 32B · 4-bit | 46±1* | local | | |
| 305 | Mistral Small 4 (119B-A6.5B) | 46±7*† | api $0.26/M +local | | |
| 306 | Gemma 2 9B IT · 8-bit | 46±1 | local | | |
| 307 | Llama 3 70B Instruct · 2-bit | 46±1 | local | | |
| 309 | Qwen2.5 7B Instruct · 8-bit | 45±2 | local | | |
| 311 |  OpenThinker2 32B · 4-bit | 45±5* | local | | |
| 312 | GPT-4 (0613) | 45±1 | api $37.50/M | | |
| 313 | Qwen2 72B Instruct · 4-bit | 45±1 | local (slow) | | |
| 314 | Nova Lite | 45±1* | api $0.10/M | | |
| 315 | Granite 3.3 8B Instruct · 8-bit | 45±4*† | local | | |
| 316 | LFM2 1.2B · 4-bit | 45±4*† | local | | |
| 318 | Command R Plus (04-2024) · 2-bit | 44±1 | local (slow) | | |
| 319 | Phi-4 14B · 8-bit | 44±1 | local | | |
| 320 | OLMo 2 32B Instruct (0325) · 4-bit | 44±1 | local | | |
| 321 | Granite 3.2 8B Instruct · 8-bit | 44±4*† | local | | |
| 322 | EXAONE 3.5 2.4B Instruct · 8-bit | 44±3*† | local | | |
| 323 | Qwen3 1.7B · 8-bit | 44±2 | local | | |
| 324 | Qwen2.5 3B Instruct · 8-bit | 44±3* | local | | |
| 325 | Qwen2 7B Instruct · 8-bit | 44±3* | local | | |
| 326 | Command R (08-2024) · 4-bit | 44±1* | local | | |
| 327 | Ministral 3 14B · 8-bit | 43±3* | local | | |
| 328 | LFM2 700M · 4-bit | 43±4*† | local | | |
| 329 | Qwen1.5 110B Chat · 2-bit | 43±1 | local (slow) | | |
| 330 | Qwen3-VL 2B Instruct · 8-bit | 43±2* | local | | |
| 331 | Nova Micro | 42±1* | api $0.061/M | | |
| 333 |  Jamba 1.5 Mini · 4-bit | 42±1 | local (slow) | | |
| 334 |  SmolLM3 3B · 8-bit | 42±4* | local | | |
| 335 | EXAONE 4.0 1.2B · 8-bit | 42±3*† | local | | |
| 337 | Ministral 8B (2410) · 16-bit | 42±1 | local | | |
| 338 | Ministral 3 8B · 8-bit | 42±6*† | local | | |
| 339 | OLMo 2 7B Instruct (1124) · 8-bit | 42±4*† | local | | |
| 340 | Qwen1.5 72B Chat · 2-bit | 41±1 | local | | |
| 342 | Aya Expanse 8B · 8-bit | 41±1* | local | | |
| 345 | Llama 3.1 Tulu 3 8B · 8-bit | 40±1 | local | | |
| 346 | GPT-3.5 Turbo (0125) | 40±1 | api $0.75/M | | |
| 347 | Command R (03-2024) · 4-bit | 40±1 | local | | |
| 350 |  Xwin-LM 70B v0.1 · 2-bit | 39±2* | local | | |
| 351 | Llama 3 8B Instruct · 8-bit | 39±1 | local | | |
| 353 | LFM2 350M · 4-bit | 39±4*† | local | | |
| 355 | Llama 3.1 8B Instruct · 8-bit | 39±1 | local | | |
| 356 | Granite 3.1 8B Instruct · 8-bit | 39±1 | local | | |
| 357 | Yi 1.5 34B Chat · 4-bit | 39±1* | local | | |
| 359 | Gemma 3 1B · 4-bit | 38±2 | local | | |
| 360 | Qwen3 0.6B · 8-bit | 38±3* | local | | |
| 361 | Ministral 3 14B Reasoning 2512 · 4-bit | 38±3* | local | | |
| 362 | Qwen1.5 32B Chat · 4-bit | 38±1* | local | | |
| 363 | Phi-3-medium-4k-instruct · 8-bit | 38±1 | local | | |
| 364 | Gemma 2 2B IT · 8-bit | 37±1* | local | | |
| 365 | Phi-3.5-vision-instruct · 16-bit | 37±8* | local | | |
| 366 | Mixtral 8x7B Instruct v0.1 · 4-bit | 37±1 | local | | |
| 369 |  InternLM2.5 20B Chat · 8-bit | 37±1* | local | | |
| 370 | InternLM2 Chat 20B · 8-bit | 36±2* | local | | |
| 371 | Mistral 7B Instruct v0.3 · 8-bit | 36±2* | local | | |
| 372 | Yi 34B Chat · 4-bit | 36±1 | local | | |
| 373 | Qwen1.5 14B Chat · 8-bit | 36±1 | local | | |
| 374 | Granite 3.3 2B Instruct · 8-bit | 36±3*† | local | | |
| 375 | WizardLM 70B V1.0 · 4-bit | 35±1 | local (slow) | | |
| 376 |  OpenChat 3.5 0106 · 8-bit | 35±1* | local | | |
| 377 | Tulu 2 DPO 70B · 4-bit | 35±1 | local (slow) | | |
| 378 | Granite 3.1 3B-A800M Instruct · 8-bit | 35±5*† | local | | |
| 380 | Granite 3.0 8B Instruct · 8-bit | 35±1* | local | | |
| 381 | OpenChat 3.5 · 8-bit | 35±1* | local | | |
| 382 | DeepSeek LLM 67B Chat · 2-bit | 35±1 | local | | |
| 384 | Gemma 4 E4B · 8-bit | 35±2 | local | | |
| 385 | Gemma 1.1 7B IT · 8-bit | 35±1 | local | | |
| 386 | Granite 3.1 2B Instruct · 8-bit | 35±1 | local | | |
| 387 | Phi-3-small-8k-instruct · 16-bit | 34±1 | local | | |
| 388 | Starling-LM 7B beta · 8-bit | 34±1 | local | | |
| 389 | Ministral 3 3B · 8-bit | 34±5* | local | | |
| 390 |  Deita 7B v1.0 · 4-bit | 34±2* | local (slow) | | |
| 391 | Granite 4.0 H 350M · 8-bit | 34±4*† | local | | |
| 392 |  Vicuna 33B v1.3 · 4-bit | 34±1 | local | | |
| 393 |  OpenHermes 2.5 Mistral 7B · 8-bit | 34±1 | local | | |
| 394 | Granite 3.2 2B Instruct · 8-bit | 34±3*† | local | | |
| 395 | Llama 3.2 3B Instruct · 8-bit | 33±1 | local | | |
| 396 | Llama 2 70B Chat · 4-bit | 33±1 | local (slow) | | |
| 397 | OLMo 2 1B Instruct (0425) · 16-bit | 33±4*† | local | | |
| 398 |  Starling-LM 7B alpha · 8-bit | 33±1 | local | | |
| 399 | OpenThinker3 7B · 16-bit | 33±3* | local | | |
| 400 | Nous Hermes 2 Mixtral 8x7B DPO · 4-bit | 33±1* | local | | |
| 401 | Gemma 3 270M IT · 8-bit | 32±5*† | local | | |
| 402 | Granite 3.0 2B Instruct · 8-bit | 32±1* | local | | |
| 403 | SmolLM2 360M Instruct · 4-bit | 32±5*† | local | | |
| 404 | QwQ-32B-Preview · 4-bit | 31±1* | local | | |
| 405 | WizardLM 13B V1.2 · 4-bit | 31±1 | local | | |
| 406 | Mistral 7B Instruct v0.2 · 8-bit | 31±1 | local | | |
| 407 | Tulu 2 DPO 13B · 16-bit | 31±2* | local | | |
| 408 | Granite 3.1 1B-A400M Instruct · 8-bit | 31±5*† | local | | |
| 409 | Granite 3.0 3B-A800M Instruct · 8-bit | 31±4*† | local | | |
| 410 | SOLAR 10.7B Instruct v1.0 · 8-bit | 31±1* | local | | |
| 411 | Qwen3.5 0.8B · 8-bit | 31±4*† | local | | |
| 412 |  Dolphin 2.2.1 Mistral 7B · 8-bit | 31±1 | local | | |
| 413 | Qwen1.5 7B Chat · 8-bit | 31±1 | local | | |
| 415 | Qwen2.5 0.5B Instruct · 8-bit | 30±3* | local | | |
| 417 | Qwen-14B-Chat · 16-bit | 30±1 | local | | |
| 418 | Phi-3-mini-4k-instruct (June 2024 update) · 8-bit | 30±1* | local | | |
| 419 | Vicuna 13B v1.5 · 8-bit | 30±1 | local | | |
| 420 | Granite 4.0 350M · 16-bit | 30±4*† | local | | |
| 421 | Llama 2 13B Chat · 8-bit | 29±1 | local | | |
| 422 | Code Llama 34B Instruct · 4-bit | 29±1* | local | | |
| 423 | Zephyr 7B beta · 8-bit | 29±1 | local | | |
| 424 | Tulu 2 DPO 7B · 16-bit | 29±2* | local | | |
| 425 | Phi-3-mini-128k-instruct (Apr 2024) · 16-bit | 29±1 | local | | |
| 426 | Gemma 7B IT · 8-bit | 29±1 | local | | |
| 427 | Phi-3-mini-4k-instruct (Apr 2024) · 4-bit | 29±1* | local | | |
| 428 | Zephyr 7B alpha · 8-bit | 28±1 | local | | |
| 430 | Gemma 4 E2B · 8-bit | 28±3 | local | | |
| 431 | Llama 3.2 1B Instruct · 4-bit | 27±1 | local | +1 | |
| 432 | Code Llama 70B Instruct · 2-bit | 27±1* | local | | |
| 434 | Vicuna 7B v1.5 · 8-bit | 27±1 | local | | |
| 435 | Gemma 1.1 2B IT · 8-bit | 27±1 | local | | |
| 436 | SmolLM2 1.7B Instruct · 8-bit | 27±1 | local | | |
| 437 | Granite 3.0 1B-A400M Instruct · 8-bit | 26±4*† | local | | |
| 438 | Qwen2.5 1.5B Instruct · 8-bit | 26±3 | local | | |
| 439 | Mistral 7B Instruct v0.1 · 8-bit | 26±1* | local | | |
| 440 | Llama 2 7B Chat · 8-bit | 25±1 | local | | |
| 441 |  Ghost 7B Alpha · 16-bit | 25±2* | local | | |
| 442 | Nous-Hermes-13b · 16-bit | 25±2* | local | | |
| 443 | Qwen1.5 4B Chat · 8-bit | 24±1* | local | | |
| 444 | Gemma 2B IT · 8-bit | 23±1 | local | | |
| 445 | OLMo 7B Instruct · 16-bit | 21±1* | local | | |
| 450 |  Falcon 40B Instruct · 4-bit | 19±1 | local | | |
| 451 |  ChatGLM3-6B · 4-bit | 19±1* | local | | |
| 452 | Falcon 7B Instruct · 8-bit | 18±2* | local | | |
| 454 | ChatGLM2-6B · 4-bit | 16±1 | local | | |
| 455 | Qwen1.5 1.8B Chat · 4-bit | 16±5* | local | | |
| 456 | Phi-2 (2.7B) · 4-bit | 14±2* | local | | |
| 457 |  OpenAssistant SFT-4 Pythia 12B · 4-bit | 14±1 | local | | |
| 459 | ChatGLM-6B · 4-bit | 12±1 | local | | |
| 460 | FastChat-T5 3B · 4-bit | 10±1 | local | | |
| 463 |  StableLM-Tuned-Alpha 7B · 8-bit | 6±1 | local | | |
| — | Hermes 3 Llama 3.1 70B · q4_0 | not yet calibrated | local (slow) | | |
| — | Hermes 3 Llama 3.1 8B · q8_0 | not yet calibrated | local | | |
| — | Phi-3.5-mini-instruct · q8_0 | not yet calibrated | local | | |
| — | Phi-4-mini 3.8B · q8_0 | not yet calibrated | local | | |
| — | Qwen2 0.5B Instruct · q8_0 | not yet calibrated | local | | |
| — | Qwen2 1.5B Instruct · q8_0 | not yet calibrated | local | |
+ 102 more don't fit your machine or budget · show them
+ 68 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them
472 of 588 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