Reasoning
DTBench Leaderboard
DTBench is a set of handcrafted, expert-validated multiple-choice questions on decision theory — Newcomb-like problems and related puzzles — built by the Conceptual Reasoning Index project with Anthropic. It tests whether a model can follow an abstract argument rather than recall a fact.
Source: epoch67 open models ranked+143 proprietaryData through Sep 2026
Open models ranked on DTBench
# shows rank among open models / rank overall (including proprietary).
| # | Model | Score |
|---|---|---|
| 1 / 44 | DeepSeek V4 Pro 0813 · 1650.5B | 93.9% |
| 2 / 46 | GLM 5.2 · 753.3B | 93.6% |
| 3 / 57 | Kimi K3 · 2779.9B | 91.2% |
| 4 / 58 | DeepSeek V4 Flash 0731 · 304.2B | 90.9% |
| 5 / 60 | Kimi K2.6 · 1026.9B | 90.9% |
| 6 / 61 | DeepSeek V4 Pro · 1598.8B | 90.7% |
| 7 / 68 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 90.1% |
| 8 / 71 | DeepSeek V4.1 Flash · 763.2B | 89.9% |
| 9 / 76 | Qwen3.8 27B · 27.8B | 88.0% |
| 10 / 77 | DeepSeek V3.2 Exp · 685.4B | 87.7% |
| 11 / 79 | GLM 5.3 · 753.3B | 87.7% |
| 12 / 80 | Inkling · 952.4B | 87.5% |
| 13 / 81 | Qwen3.5 397B A17B · 403.4B | 87.5% |
| 14 / 85 | DeepSeek V4 Flash · 290.9B | 86.4% |
| 15 / 86 | DeepSeek V3.2 · 685.4B | 85.6% |
| 16 / 89 | MiMo V2.5 Pro · 1023.2B | 84.5% |
| 17 / 90 | Qwen3.5 122B A10B · 125.1B | 84.3% |
| 18 / 97 | DeepSeek V3.1 · 684.5B | 82.7% |
| 19 / 98 | Gemma 4 31B IT · 31.3B | 82.7% |
| 20 / 101 | Qwen3.5 27B · 27.8B | 82.4% |
| 21 / 105 | DeepSeek V3.1 Terminus · 684.5B | 81.3% |
| 22 / 110 | Qwen3 235B A22B Thinking 2507 · 235.1B | 80.3% |
| 23 / 114 | Qwen3.5 35B A3B · 36.0B | 80.0% |
| 24 / 116 | MiniMax M3 · 427.0B | 78.9% |
| 25 / 118 | Qwen3 235B A22B Instruct 2507 · 235.1B | 78.4% |
| 26 / 119 | Qwen3.6 27B · 27.8B | 78.1% |
| 27 / 125 | GPT OSS 120B · 116.8B | 76.3% |
| 28 / 126 | Qwen3 235B A22B · 235.1B | 75.7% |
| 29 / 128 | Gemma 4 26B A4B IT · 25.8B | 74.9% |
| 30 / 130 | Qwen3.6 35B A3B · 36.0B | 73.9% |
| 31 / 133 | Qwen3.5 9B · 9.7B | 71.2% |
| 32 / 135 | Qwen3 30B A3B Thinking 2507 · 30.5B | 69.3% |
| 33 / 140 | GPT OSS 20B · 20.9B | 68.0% |
| 34 / 144 | Qwen3 32B · 32.8B | 67.5% |
| 35 / 145 | Qwen3 30B A3B Instruct 2507 · 30.5B | 67.2% |
| 36 / 147 | Mistral Large 3 675B Instruct 2512 · 675B | 65.1% |
| 37 / 148 | DeepSeek v3 0324 · 684.5B | 64.8% |
| 38 / 150 | Qwen3 14B · 14.8B | 64.0% |
| 39 / 152 | Qwen2.5 72B Instruct · 72.7B | 62.9% |
| 40 / 158 | Llama 4 Maverick 17B 128E Instruct · 401.6B | 61.9% |
| 41 / 161 | Llama 3.1 405B Instruct · 405.9B | 61.4% |
| 42 / 162 | C4ai Command A 03 2025 · 111.1B | 61.3% |
| 43 / 163 | Magistral Small 2509 · 24.0B | 61.3% |
| 44 / 164 | Mistral Large Instruct 2407 · 122.6B | 61.2% |
| 45 / 165 | Mistral Large Instruct 2411 · 122.6B | 60.8% |
| 46 / 166 | Qwen3 30B A3B · 30.5B | 60.3% |
| 47 / 167 | Llama 3.1 70B Instruct · 70.6B | 60.0% |
| 48 / 168 | Mistral Small 3.2 24B Instruct 2506 · 24.0B | 59.9% |
| 49 / 169 | Qwen3 8B · 8.2B | 59.7% |
| 50 / 170 | Llama 3.3 70B Instruct · 70.6B | 59.5% |
| 51 / 172 | Mistral Small 3.1 24B Instruct 2503 · 24.0B | 58.6% |
| 52 / 173 | Llama 4 Scout 17B 16E Instruct · 108.6B | 57.9% |
| 53 / 177 | Mixtral 8x22B Instruct v0.1 · 140.6B | 55.1% |
| 54 / 178 | C4ai Command R Plus 08 2024 · 103.8B | 54.9% |
| 55 / 180 | Meta Llama 3 70B Instruct · 70.6B | 54.2% |
| 56 / 186 | Gemma 3 27B IT · 27.4B | 52.5% |
| 57 / 192 | Gemma 3 4B IT · 4.3B | 50.9% |
| 58 / 193 | Llama 3.1 8B Instruct · 8.0B | 50.9% |
| 59 / 196 | Mixtral 8x7B Instruct v0.1 · 46.7B | 49.6% |
| 60 / 197 | Mistral Small Instruct 2409 · 22.2B | 49.5% |
| 61 / 198 | Gemma 3 12B IT · 12.2B | 48.8% |
| 62 / 199 | Mistral Nemo Instruct 2407 · 12.2B | 48.6% |
| 63 / 202 | Qwen2.5 7B Instruct · 7.6B | 47.7% |
| 64 / 203 | C4ai Command R 08 2024 · 32.3B | 46.4% |
| 65 / 207 | Meta Llama 3 8B Instruct · 8.0B | 43.9% |
| 66 / 209 | Llama 2 13B Chat HF · 13.0B | 42.2% |
| 67 / 210 | Llama 2 70B Chat HF · 69.0B | 41.6% |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Gemma 3 4B IT, 4B, score 50.9% — on the efficiency frontier (best score at its size or smaller).
- Qwen3 8B, 8B, score 59.7% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.5 9B, 10B, score 71.2% — on the efficiency frontier (best score at its size or smaller).
- Gemma 4 26B A4B IT, 26B, score 74.9% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.8 27B, 28B, score 88.0% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Flash 0731, 304B, score 90.9% — on the efficiency frontier (best score at its size or smaller).
- GLM 5.2, 753B, score 93.6% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Pro 0813, 1.7T, score 93.9% — on the efficiency frontier (best score at its size or smaller).
DTBench: frequently asked questions
- What is the best open LLM on DTBench?
- DeepSeek V4 Pro 0813 is the top open model on DTBench, scoring 93.9%. Among all models tested — including proprietary ones — it ranks #44. The top model overall is Claude Fable 5 Max (Anthropic) at 98.4%.
- What's the best DTBench model you can run on a 24 GB GPU?
- Qwen3.8 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 15 GB), scoring 88.0% on DTBench.
- What's the best DTBench model you can run on a 12 GB GPU?
- Qwen3.5 9B is the highest-scoring open model that fits in 12 GB at 4-bit quantization (about 5 GB), scoring 71.2% on DTBench.
- Can open models match proprietary models on DTBench?
- Not quite on DTBench: the strongest proprietary model (Claude Fable 5 Max) scores 98.4%, ahead of the best open model (DeepSeek V4 Pro 0813) at 93.9% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.