LLM benchmark scores, with the gaps filled
benchgap is an LLM benchmark leaderboard that fills in the missing scores. Most models are only ever run on a handful of benchmarks, so benchgap calibrates benchmarks against each other on the models measured on both, then estimates each missing score with its cross-validated error and a confidence level. Measured and estimated scores are always marked apart.
As of 2026-11-21: 612 models, 240 benchmark versions, 8226 measured and 27521 estimated scores (5081 high, 9813 medium and 12627 low confidence). Measured scores come from public leaderboards (artificial-analysis, published, vals-ai).
AA-HLE leaderboard
As of 2026-11-21, the highest measured score on AA-HLE is 74.3% by Claude Fable 5.2. 97 more models have estimated scores, calibrated from the benchmarks they were measured on.
Measured scores: benchlm.ai.
| # | Model | Score | Source |
|---|---|---|---|
| 1 | Claude Fable 5.2 | 74.3% | measured |
| 2 | Claude Opus 5.5 | 61.4% | measured |
| 3 | GPT-6.2 Astra | 61.4% | estimated ± 3.9 pp, medium confidence |
| 4 | Claude Fable 5.1 | 59.1% | measured |
| 5 | Gemini 4 Neon | 59.0% | estimated ± 9.4 pp, medium confidence |
| 6 | Claude Mythos 5 | 57.9% | estimated ± 4.6 pp, high confidence |
| 7 | Gemini 4 Argon | 57.1% | measured |
| 8 | GPT-6.2 Sol | 56.1% | estimated ± 3.9 pp, high confidence |
| 9 | Claude Fable 5 | 55.5% | measured |
| 10 | Claude Sonnet 5.5 | 55.0% | measured |
| 11 | Claude Opus 5 | 54.9% | measured |
| 12 | GPT-6 Astra | 54.7% | measured |
| 13 | Gemini 4 Hydrogen | 53.7% | estimated ± 9.4 pp, medium confidence |
| 14 | Sakana Fugu | 53.7% | estimated ± 3.9 pp, high confidence |
| 15 | Sakana Fugu-Ultra | 53.7% | estimated ± 3.9 pp, high confidence |
| 16 | Muse Wonder Full | 53.6% | measured |
| 17 | GPT-6.1 Sol | 52.9% | measured |
| 18 | Gemini 4 Carbon | 52.8% | estimated ± 6.0 pp, medium confidence |
| 19 | Kimi K3.1 Pro | 52.5% | measured |
| 20 | Qwen4 Max | 50.5% | estimated ± 4.6 pp, high confidence |
| 21 | Muse Wonder | 50.1% | measured |
| 22 | GPT-5.6 Sol | 49.5% | measured |
| 23 | MiMo-V2.6-Pro | 49.4% | measured |
| 24 | Claude Opus 4.8 | 48.7% | measured |
| 25 | Muse Spark 1.3 | 48.7% | measured |
| 26 | MiniMax M3.1 Flash | 48.0% | measured |
| 27 | Gemini 3.7 Flash | 47.9% | measured |
| 28 | GPT-6 Sol | 47.9% | measured |
| 29 | Gemini 3.8 Flash | 47.8% | measured |
| 30 | Pareto 26.9 | 47.8% | estimated ± 4.6 pp, high confidence |
| 31 | dots3-note Preview | 47.7% | estimated ± 6.0 pp, medium confidence |
| 32 | Hy4 | 47.1% | estimated ± 6.5 pp, medium confidence |
| 33 | Gemini 3.1 Pro | 47.0% | measured |
| 34 | Kimi K3 | 46.9% | measured |
| 35 | Step 5 Preview | 46.5% | measured |
| 36 | GPT-6.1 Terra | 46.2% | measured |
| 37 | Muse Spark 1.1 | 46.2% | measured |
| 38 | GPT-5.5 | 45.8% | measured |
| 39 | Sage-A1.5 | 45.7% | estimated ± 5.8 pp, medium confidence |
| 40 | Muse Spark 1.2 | 45.5% | measured |
| 41 | ERNIE 5.5 Flash Preview | 45.2% | estimated ± 4.6 pp, high confidence |
| 42 | Hy4 flash | 44.8% | estimated ± 6.5 pp, medium confidence |
| 43 | Claude Haiku 5.5 | 44.4% | measured |
| 44 | Inkling 2 Small | 44.1% | measured |
| 45 | Muse Spark Open | 43.9% | measured |
| 46 | GPT-5.4 | 43.7% | measured |
| 47 | Grok 4.7 | 43.1% | measured |
| 48 | Qwen3.8 Max Preview | 43.1% | measured |
| 49 | Ornith-1.5-397B | 43.1% | estimated ± 3.9 pp, high confidence |
| 50 | GPT-5.6 Terra | 42.9% | measured |
| 51 | Grok 4.6 | 42.9% | measured |
| 52 | Gemini 3.5 Flash | 42.7% | measured |
| 53 | Grok 4.5 | 42.7% | measured |
| 54 | GPT-5.3 Codex | 42.5% | measured |
| 55 | Qwen3.8 Max | 42.4% | estimated ± 3.9 pp, high confidence |
| 56 | Claude Opus 4.7 | 42.3% | measured |
| 57 | GLM-5.3 | 42.3% | measured |
| 58 | GPT-5.5 Pro | 42.0% | estimated ± 4.6 pp, high confidence |
| 59 | Pareto 26.10 Preview | 41.8% | estimated ± 3.9 pp, high confidence |
| 60 | GPT-5.4 Pro | 41.7% | estimated ± 4.6 pp, high confidence |
| 61 | Hy4 preview | 41.5% | estimated ± 3.9 pp, high confidence |
| 62 | Claude Sonnet 5 | 41.3% | measured |
| 63 | GLM-5.2 | 41.1% | measured |
| 64 | DeepSeek V4 Pro | 41.0% | measured |
| 65 | Gemini 3.6 Flash | 40.8% | measured |
| 66 | Muse Spark | 40.7% | measured |
| 67 | Qwen3.7 Max | 40.5% | measured |
| 68 | Agents-A1 | 40.2% | estimated ± 6.5 pp, medium confidence |
| 69 | Claude Opus 4.6 (Adaptive) | 39.9% | measured |
| 70 | Gemini 3 Pro | 39.7% | measured |
| 71 | GPT-5.6 Luna | 39.5% | measured |
| 72 | Ling 3.1 Flash | 39.4% | measured |
| 73 | Muse Ember | 39.4% | estimated ± 9.4 pp, medium confidence |
| 74 | DeepSeek V4.1 Flash 1017 | 39.2% | measured |
| 75 | MiniMax M3 | 39.0% | measured |
| 76 | DeepSeek V4 Flash 0731 | 38.6% | measured |
| 77 | GPT-6 Luna | 38.5% | measured |
| 78 | Qwen3.8-Flash-Next | 38.0% | measured |
| 79 | GPT-5.2 | 37.7% | measured |
| 80 | Qwen3.8-Omni-Flash | 37.6% | estimated ± 3.9 pp, high confidence |
| 81 | Kimi K2.6 | 37.5% | measured |
| 82 | Grok 4.3 | 37.2% | measured |
| 83 | Mistral Large 4 | 36.9% | estimated ± 5.8 pp, medium confidence |
| 84 | Particle-1 | 36.0% | estimated ± 3.9 pp, high confidence |
| 85 | Beam 1.0 501B-A23B FP8 | 35.7% | estimated ± 3.9 pp, high confidence |
| 86 | GPT-5.2-Codex | 35.7% | measured |
| 87 | MiMo-V2.5-Pro | 35.7% | measured |
| 88 | Qwen3.7 Plus | 35.6% | measured |
| 89 | MiMo-V2.6-Flash | 35.1% | measured |
| 90 | Kimi K2.7 Code | 35.0% | measured |
| 91 | Interfaze Beta | 34.7% | estimated ± 3.9 pp, high confidence |
| 92 | Beam 1.0 501B-A23B | 34.7% | measured |
| 93 | Apodex 1.1 | 34.1% | measured |
| 94 | Apodex 1.1 Mini | 34.1% | measured |
| 95 | Qwen3.8-27B | 33.9% | measured |
| 96 | Hy3 | 33.5% | measured |
| 97 | Hy3 Preview | 33.5% | measured |
| 98 | Inkling-Small | 33.3% | measured |
| 99 | Beam 1.0 501B-A23B NVFP4 | 33.3% | estimated ± 3.9 pp, high confidence |
| 100 | Mercury 3 | 33.1% | estimated ± 3.9 pp, high confidence |
| 101 | Ornith-1.5-35B-A3B | 33.1% | estimated ± 3.9 pp, high confidence |
| 102 | Nemotron 3.5 Bolt 74B-A9B BF16 | 32.2% | estimated ± 3.9 pp, high confidence |
| 103 | Inkling | 31.9% | measured |
| 104 | Grok 4.20 | 31.5% | estimated ± 3.9 pp, high confidence |
| 105 | OrcaCyber-Nano-1.0 | 31.0% | estimated ± 3.9 pp, high confidence |
| 106 | Qwen 3.6 Max (preview) | 30.8% | measured |
| 107 | Kimi K2.5 | 30.7% | measured |
| 108 | Nemotron 3.5 Super VL 120B A12B | 30.7% | measured |
| 109 | Nemotron 3.5 Super VL 120B A12B NVFP4 | 30.7% | measured |
| 110 | MiMo-V2-Pro | 30.4% | measured |
| 111 | Claude Opus 4.5 Thinking | 30.1% | measured |
| 112 | GLM-5.1 | 30.1% | measured |
| 113 | GLM-5.3-Flash | 29.9% | estimated ± 6.2 pp, medium confidence |
| 114 | Solar Pro 4 | 29.8% | measured |
| 115 | Claude Sonnet 4.5 Thinking | 29.6% | estimated ± 6.0 pp, medium confidence |
| 116 | A.X K2 | 29.6% | measured |
| 117 | MiniMax M2.7 | 29.6% | measured |
| 118 | GLM-5 | 29.3% | measured |
| 119 | Qwen3.5 397B | 29.0% | measured |
| 120 | GPT-5 | 28.5% | measured |
| 121 | GPT-5.1 | 28.5% | measured |
| 122 | Nemotron 3 Ultra | 28.4% | measured |
| 123 | GPT-5.4 nano | 28.3% | measured |
| 124 | GPT-5.4 mini | 28.1% | measured |
| 125 | GLM-5-Turbo | 27.8% | measured |
| 126 | Qwen3.6 Plus | 27.8% | measured |
| 127 | Inkling 2 Mini | 27.7% | estimated ± 3.9 pp, high confidence |
| 128 | GLM-4.7 | 27.4% | measured |
| 129 | Ornith-1.5-9B | 27.3% | estimated ± 3.9 pp, high confidence |
| 130 | Solar Open 2 | 27.1% | estimated ± 3.9 pp, high confidence |
| 131 | Grok 4 | 26.7% | measured |
| 132 | Seed 2.1 Pro | 26.6% | estimated ± 7.6 pp, low confidence |
| 133 | Ternary Bonsai 2 27B | 26.2% | estimated ± 3.9 pp, high confidence |
| 134 | Seed 2.1 Turbo | 26.1% | estimated ± 7.6 pp, low confidence |
| 135 | GPT-5.1-Codex | 25.7% | measured |
| 136 | GPT-5.1-Codex-Max | 25.7% | measured |
| 137 | Ripple-65B Preview | 25.2% | estimated ± 3.9 pp, high confidence |
| 138 | Qwen3.5-122B-A10B | 25.2% | measured |
| 139 | Nemotron 3.5 Bolt 74B-A9B | 24.7% | measured |
| 140 | o3-pro | 24.0% | estimated ± 2.7 pp, high confidence |
| 141 | Qwen3.5-27B | 23.9% | measured |
| 142 | Ling 3.0 Flash | 23.7% | measured |
| 143 | MAI-Thinking-1 | 23.6% | estimated ± 3.9 pp, high confidence |
| 144 | Gemma 4 31B | 23.6% | measured |
| 145 | Ling 3.0 Flash FP8 | 23.3% | estimated ± 3.9 pp, high confidence |
| 146 | Qwen3.6-27B | 23.1% | measured |
| 147 | Gemini 2.5 Pro | 22.5% | measured |
| 148 | Claude 4.1 Opus | 22.4% | estimated ± 9.4 pp, medium confidence |
| 149 | Qwen3.5 Flash | 22.4% | estimated ± 6.2 pp, medium confidence |
| 150 | Qwen3.6-35B-A3B | 22.2% | measured |
| 151 | MiMo-V2-Omni | 22.1% | measured |
| 152 | Ling 3.0 Flash VL | 22.0% | measured |
| 153 | Muse Glimmer 30B | 22.0% | measured |
| 154 | Step 3.7 Flash | 21.4% | measured |
| 155 | Qwen3.5-35B-A3B | 21.0% | measured |
| 156 | MiMo-V2.5 | 20.2% | estimated ± 6.2 pp, medium confidence |
| 157 | o3 | 20.1% | measured |
| 158 | Claude Sonnet 4.5 | 19.9% | estimated ± 5.7 pp, medium confidence |
| 159 | GPT-OSS 120B | 19.6% | measured |
| 160 | Gemini 3.1 Flash-Lite | 19.4% | estimated ± 6.2 pp, medium confidence |
| 161 | Gemma 4 26B A4B | 19.3% | measured |
| 162 | Grok 4.1 Fast (Reasoning) | 19.3% | measured |
| 163 | Claude Opus 4.6 | 19.1% | measured |
| 164 | Grok 4 Fast (Reasoning) | 19.1% | measured |
| 165 | Gemini 3.5 Flash-Lite | 18.8% | measured |
| 166 | Quasar 438B | 18.7% | measured |
| 167 | K-EXAONE 2.0 | 18.6% | measured |
| 168 | GLM-5V-Turbo | 17.1% | measured |
| 169 | DeepSeek-R1-0528 | 17.0% | estimated ± 5.7 pp, medium confidence |
| 170 | o4-mini | 16.5% | measured |
| 171 | Grok 3 Mini | 16.3% | estimated ± 5.7 pp, medium confidence |
| 172 | DeepSeek-R1 | 15.8% | measured |
| 173 | Trinity-Large-Preview | 15.8% | measured |
| 174 | Trinity-Large-Thinking | 15.8% | measured |
| 175 | Gemma 4 12B | 15.7% | measured |
| 176 | Nemotron 3 Super 100B | 15.2% | estimated ± 5.7 pp, medium confidence |
| 177 | Spark-X2.6-4B | 15.0% | estimated ± 3.9 pp, high confidence |
| 178 | o1-pro | 15.0% | estimated ± 5.7 pp, medium confidence |
| 179 | Gemini 3 Flash | 15.0% | measured |
| 180 | K-Exaone | 13.9% | measured |
| 181 | Mistral Medium 3.5 128B | 13.8% | measured |
| 182 | Qwen3 235B 2507 | 13.7% | estimated ± 5.7 pp, medium confidence |
| 183 | Claude Sonnet 4.6 | 13.3% | measured |
| 184 | Claude Opus 4.5 | 13.2% | measured |
| 185 | LFM2.5-24B-A2B | 12.6% | estimated ± 3.9 pp, high confidence |
| 186 | Claude 4 Opus Thinking | 12.3% | measured |
| 187 | o1-preview | 12.1% | estimated ± 2.7 pp, high confidence |
| 188 | Grok 3 [Beta] | 12.0% | estimated ± 5.7 pp, medium confidence |
| 189 | Command A+ | 12.0% | measured |
| 190 | Qwen3 Max | 11.9% | measured |
| 191 | GLM-4.7-Flash | 11.9% | estimated ± 5.7 pp, medium confidence |
| 192 | Mercury 2.5 | 11.8% | measured |
| 193 | DeepSeek V3.2 | 11.2% | measured |
| 194 | Granite 4.2 30B | 11.2% | measured |
| 195 | North Mini Code | 11.1% | measured |
| 196 | GPT-OSS 20B | 11.0% | measured |
| 197 | Sarvam 105B | 11.0% | measured |
| 198 | Nemotron 3.5 Lightning 30B A3B NVFP4 | 10.6% | measured |
| 199 | Solar Pro 3 | 10.3% | measured |
| 200 | ZAYA1-8B | 10.1% | estimated ± 3.9 pp, high confidence |
| 201 | Mistral Small 4 | 9.9% | measured |
| 202 | Qwen3.5-4B | 9.9% | measured |
| 203 | Magistral Medium | 9.8% | measured |
| 204 | Claude 3.7 Sonnet Thinking | 9.7% | measured |
| 205 | Granite 4.2 8B | 9.7% | measured |
| 206 | Claude Haiku 4.5 | 9.3% | estimated ± 6.2 pp, medium confidence |
| 207 | GLM-4.5 | 9.3% | estimated ± 6.2 pp, medium confidence |
| 208 | Qwen3 235B-A22B (Reasoning) | 9.3% | estimated ± 5.7 pp, medium confidence |
| 209 | Ling 3.0 Tiny | 9.3% | measured |
| 210 | LongCat-Flash-Lite-Sparse | 9.1% | estimated ± 3.9 pp, high confidence |
| 211 | MiniCPM5-2B | 8.9% | measured |
| 212 | MiMo-V2-Flash | 8.6% | measured |
| 213 | GPT-4.5 | 8.5% | estimated ± 5.7 pp, medium confidence |
| 214 | Phi-4-reasoning-plus | 8.2% | estimated ± 5.7 pp, medium confidence |
| 215 | Grok Code Fast 1 | 8.0% | measured |
| 216 | Qwen3 32B (Reasoning) | 7.9% | estimated ± 5.7 pp, medium confidence |
| 217 | o3-mini | 7.9% | measured |
| 218 | Magistral Small | 7.6% | measured |
| 219 | Qwen3-Omni-30B-A3B-Thinking | 7.5% | measured |
| 220 | Sarvam 30B | 7.5% | measured |
| 221 | Kimi K2 | 7.4% | measured |
| 222 | Nemotron Ultra 253B | 7.4% | measured |
| 223 | QwQ-32B | 7.4% | measured |
| 224 | GLM-4.5-Air | 7.0% | measured |
| 225 | o1 | 7.0% | measured |
| 226 | LFM2.5-8B-A1B | 6.9% | measured |
| 227 | Celeris-1 | 6.8% | measured |
| 228 | DeepSeek V3.1 | 6.7% | measured |
| 229 | Granite 4.2 3B | 6.6% | measured |
| 230 | DeepSeek R1 Distill Llama 70B | 6.5% | estimated ± 5.7 pp, medium confidence |
| 231 | Granite-4.0-H-350M | 6.4% | measured |
| 232 | Ling 2.6 Flash | 6.3% | measured |
| 233 | Mellum2.1-12B-A2.5B-Thinking | 6.3% | estimated ± 3.9 pp, high confidence |
| 234 | Claude 4 Opus | 6.2% | measured |
| 235 | LFM2.5-2.6B | 6.2% | measured |
| 236 | MiniCPM5-1B | 6.1% | estimated ± 6.7 pp, medium confidence |
| 237 | Exaone 4.0 1.2B | 5.7% | measured |
| 238 | LFM3-1.7B | 5.7% | estimated ± 3.9 pp, high confidence |
| 239 | Qwen3 235B-A22B | 5.6% | estimated ± 5.7 pp, medium confidence |
| 240 | GLM-4.6 | 5.5% | measured |
| 241 | Granite-4.0-350M | 5.5% | measured |
| 242 | Grok 4.1 Fast | 5.1% | measured |
| 243 | LFM2.5-VL-1.6B-Extract | 5.1% | measured |
| 244 | Exaone 4.0 32B | 5.0% | measured |
| 245 | GPT-4.1 mini | 5.0% | measured |
| 246 | Granite-4.0-H-1B | 5.0% | measured |
| 247 | Phi-4 Multimodal Instruct | 5.0% | measured |
| 248 | DeepSeek-R1-0528-Qwen3-8B | 5.0% | estimated ± 5.7 pp, medium confidence |
| 249 | Llama 4 Maverick | 4.9% | measured |
| 250 | Gemma 4 E2B | 4.8% | measured |
| 251 | Nemotron 3 Nano Omni 30B A3B | 4.8% | measured |
| 252 | Laguna XS.2 | 4.8% | estimated ± 6.2 pp, medium confidence |
| 253 | DeepSeek V3 0324 | 4.7% | measured |
| 254 | Gemini 2.5 Flash | 4.7% | measured |
| 255 | Qwen2.5-Max | 4.7% | estimated ± 5.7 pp, medium confidence |
| 256 | DeepSeek R1 Distill Qwen 32B | 4.6% | measured |
| 257 | Gemini 1.5 Pro | 4.6% | measured |
| 258 | Nemotron 3 Nano 30B | 4.6% | measured |
| 259 | Qwen3-Omni-30B-A3B-Instruct | 4.6% | measured |
| 260 | Claude 3.5 Sonnet | 4.4% | estimated ± 5.7 pp, medium confidence |
| 261 | Gemma 3 27B | 4.4% | measured |
| 262 | DeepSeek R1 Distill Qwen 14B | 4.3% | estimated ± 5.7 pp, medium confidence |
| 263 | DeepSeek-V2.5 | 4.3% | estimated ± 2.7 pp, high confidence |
| 264 | Claude 4 Sonnet | 4.3% | measured |
| 265 | Gemini 2.0 Flash | 4.3% | measured |
| 266 | Llama 3.1 8B | 4.3% | measured |
| 267 | Mistral Small 3.1 | 4.3% | measured |
| 268 | Laguna M.1 | 4.3% | estimated ± 6.2 pp, low confidence |
| 269 | Claude 3.7 Sonnet | 4.2% | measured |
| 270 | Gemini 1.0 Pro | 4.2% | measured |
| 271 | GPT-4.1 | 4.2% | measured |
| 272 | GPT-4o mini | 4.2% | measured |
| 273 | Mistral Large 3 | 4.2% | measured |
| 274 | Claude 3 Haiku | 4.1% | measured |
| 275 | Mistral Medium 3 | 4.1% | measured |
| 276 | Llama 3.1 405B | 4.0% | measured |
| 277 | Devstral Small | 3.8% | measured |
| 278 | Gemma 4 E4B | 3.8% | measured |
| 279 | GPT-4.1 nano | 3.8% | measured |
| 280 | Llama 4 Scout | 3.8% | measured |
| 281 | Phi-4 | 3.8% | measured |
| 282 | Phi-4-reasoning | 3.8% | measured |
| 283 | Gemini 2.5 Flash-Lite | 3.7% | measured |
| 284 | Solar Pro 2 | 3.7% | measured |
| 285 | Claude 3.5 Haiku | 3.6% | measured |
| 286 | Llama 3.1 70B | 3.6% | measured |
| 287 | Llama 3.3 70B | 3.6% | measured |
| 288 | o1-mini | 3.6% | measured |
| 289 | Ultravox v0.6 Llama 3.3 70B | 3.6% | measured |
| 290 | Qwen2.5 Coder 32B Instruct | 3.5% | measured |
| 291 | Grok 2 | 3.4% | estimated ± 5.7 pp, medium confidence |
| 292 | Mistral Large 2 | 3.3% | measured |
| 293 | Gemini 1.5 Flash | 3.2% | measured |
| 294 | Nova Pro | 3.2% | measured |
| 295 | Mellum2-12B-A2.5B-Thinking | 3.2% | estimated ± 3.9 pp, medium confidence |
| 296 | GPT-4 Turbo | 3.1% | measured |
| 297 | Qwen3 32B | 3.1% | estimated ± 5.7 pp, medium confidence |
| 298 | ZAYA1-74B-Preview | 3.1% | estimated ± 3.9 pp, medium confidence |
| 299 | MiniMax-Text-01 | 3.0% | estimated ± 5.7 pp, medium confidence |
| 300 | DeepSeek V3 | 2.9% | measured |
| 301 | Claude 3 Opus | 2.8% | measured |
| 302 | GPT-4o | 2.4% | measured |
| 303 | Grok 2 mini | 2.2% | estimated ± 5.7 pp, medium confidence |
| 304 | Gemini 2.0 Flash-Lite | 2.1% | estimated ± 5.7 pp, medium confidence |
| 305 | Qwen2.5-VL-32B | 1.2% | estimated ± 5.7 pp, medium confidence |
| 306 | Granite-4.0-1B | 0.0% | estimated ± 5.7 pp, medium confidence |
| 307 | LFM2.5-230M | 0.0% | estimated ± 5.7 pp, medium confidence |
| 308 | LFM2.5-350M | 0.0% | estimated ± 5.7 pp, medium confidence |
| 309 | LFM2.5-VL-450M | 0.0% | estimated ± 5.7 pp, medium confidence |
| 310 | LLaDA2.2-mini | 0.0% | estimated ± 3.9 pp, medium confidence |
| 311 | Mellum2-12B-A2.5B-Instruct | 0.0% | estimated ± 3.9 pp, medium confidence |
| 312 | Phi-3.5-mini | 0.0% | estimated ± 5.7 pp, medium confidence |
| 313 | Soofi S 30B-A3B | 0.0% | estimated ± 3.9 pp, medium confidence |
All leaderboards
Agentic · terminal
- AA Terminal-Bench 2.1
- AA Terminal-Bench 4.0
- Terminal-Bench 4.0
- Terminal-Bench 2.0
- Terminal-Bench 2.1
- Terminal-Bench 3.0
- Terminal-Bench Hard
- Terminal-Bench-Science 0.1
- Terminal-Bench 4.1
- Terminal-Bench 2.1 (Vals AI)
Agentic · tools
- AA Agentic Index
- AA-AnalystAgent
- AA AutomationBench
- CWE-Bench-AA
- AA Cyber Index
- DeepsecBench-AA
- AA EnterpriseOps-Gym
- GDP.pdf
- AA Harvey LAB v1.0
- Harvey LAB-AA v1.1
- AA ITBench
- AA Tau3 Banking
- Agents' Last Exam
- APEX-Agents
- APEX-Agents-AA
- ApprenticeBench
- AutomationBench
- BFCL v4
- BrowseComp
- Claw-Eval
- CWE-bench v1
- CWE-bench
- CyberGym
- DeepPlanning
- DeepSearchQA
- DRACO
- ExploitGym
- Finance Agent v2
- GDPval-AA
- Gert Labs
- Hermes Bench
- Hermes Index
- HLE w/ tools
- JobBench
- LifeVal
- MazeBench
- MCP Atlas
- MCP-Tasks
- MKA
- OSWorld 2.1
- OSWorld 2.1 Offline
- OSWorld-Verified
- OSWorld 2.0
- PinchBench
- QwenClawBench
- ResearchClawBench
- SkillsBench
- τ²-bench results
- τ³-bench results
- Toolathlon
- Toolathlon-Verified
- VITA-Bench
- WideResearch
Coding
- AA Coding Index
- AA LiveCodeBench
- AA-SciCode
- Bug Hunt Bench
- CursorBench 3.1
- CursorBench 3.2
- CursorBench 4.0
- cursorBench41
- DeepSWE
- FrontierCode 1.1 Main
- FrontierCode 2.0
- FrontierCode 1.1 Extended
- FrontierSWE v2
- LiveCodeBench
- LiveCodeBench Pro
- LiveCodeBench v6
- NL2Repo
- OpenHarmony Bench v1.2
- OpenHarmony Bench v1.0
- PostTrainBench
- PostTrainBench v1
- PostTrainBench v1.1
- ProgramBench
- React Native Evals
- ScarfBench
- SciCode
- SWE-Atlas Codebase QnA
- SWE-bench Pro
- SWE-bench Verified
- SWE-Rebench
- SWE-Marathon
- Vals LiveCodeBench
- Vals SWE-bench
- Vibe Code Bench
- VulcanBench Frontier v4
- VulcanBench v3
Math
- AIME 2024
- AIME 2025
- AIME26
- FrontierMath v2 (Tier 4)
- FrontierMath v2 (Tiers 1-3)
- HMMT Feb 2025
- HMMT Feb 2026
- HMMT Nov 2025
- IMOAnswerBench
- MATH-500
- MMAnswerBench
Knowledge & reasoning
- AA-GPQA Diamond
- HealthBench Professional (length-adjusted, Artificial Analysis)
- HealthBench Professional (raw, Artificial Analysis)
- AA-HLE
- AA MMLU-Pro
- ARC-AGI-2
- ARC-AGI-3 Provider Adapter
- ARC-AGI-1
- ARC-AGI-3
- Artificial Analysis Intelligence Index
- BBH
- BioBasicsBench
- BioMysteryBench (human-difficult)
- BioMysteryBench (human-solvable)
- C-Eval
- CritPt
- GPQA
- GPQA Diamond
- HealthBench (raw)
- HealthBench Hard
- HealthBench (length-adjusted)
- HealthBench Professional
- HealthBench Professional (raw)
- HLE
- HLE w/o tools
- HLE-Verified
- JevBench 1.4
- LABBench2
- MedXpertQA (Text)
- MMLU
- MMLU-Pro
- MMLU-Pro (Arcee)
- MMLU-Redux
- MMMLU
- AA-Omniscience Accuracy
- SimpleQA
- STEMBench
- SuperGPQA
- Vals GPQA Diamond
- Vals MMLU-Pro
Instruction following
Multilingual
Vision & documents
- AA-MMMU-Pro
- AI2D_TEST
- BabyVision
- BabyVision w/ Python
- BenchCAD Vision2Code (no tools)
- BenchCAD Vision2Code (tools)
- CC-OCR
- Chartography (no tools)
- Chartography (tools)
- CharXiv
- CharXiv w/o tools
- CountBench
- DynaMath
- ERQA
- LVBench
- MathVision
- MathVision w/ Python
- MedXpertQA (MM)
- MLVU (M-Avg)
- MMMU
- MMMU-Pro
- MMMU-Pro w/ Python
- MMVU
- OCRBench V2
- OfficeQA
- OfficeQA Pro
- OmniDocBench 1.5
- ParseBench-1
- ParseBench-2
- PerceptionBench
- RealWorldQA
- RefCOCO (avg)
- ScreenSpot Pro
- SimpleVQA
- SWE-bench Multimodal
- Video-MME (w/o subtitle)
- Video-MME (with subtitle)
- VideoMMMU
- V*
- We-Math
- ZeroBench
- ZeroBench w/ Python
Long context
General
- HLE-MultiChoice (Decision Index)
- Artificial Analysis Decision Index v1.0
- Artificial Analysis Decision Index
- When2Call (Decision Index)
- WildGuardTest (Decision Index)
- BabaIsBench
- BinaryAudit
- BlenderWorld (text-only)
- Chess Puzzles
- CompileBench
- Cookie-Bench
- DAYJOB
- DAYJOB Healthcare
- Decision Index Full Score
- EBR-bench
- ExploitBench
- ExploitBench Internal Port
- FACTS Multimodal
- FailureSensorIQ
- FrontierPhysics
- Furniture Assembly
- Decision Index Arts
- Decision Index Knowledge
- Decision Index Language
- Decision Index Retrieval
- Decision Index Tools
- MaxxBench
- MirrorCode
- Mystery Game Puzzles
- NYT Connections
- OTelBench
- SEC-Bench Pro
- SkillsBench official leaderboard (BenchFlow)
- SpireClimb
- StealthMark