Category Score V3 leaderboard
LMSpeed Best Models for Reasoning
Compare the best AI reasoning models across abstract logic, scientific reasoning, multi-step constraints, and evidence verification benchmarks with uncertainty-aware Category Scores.
Methodology 3.0Methodology
Current answer
Among the currently visible formally ranked models, GPT-5.6 Sol has the highest position at global rank 1. Its Category Score is 61.2, with an 80% uncertainty range of ±8.7. 64 visible models have a formal rank. This result applies only to this score run.
Available leaderboard data
- Models shown
- 100
- Formally ranked models
- 64
- Benchmark columns
- 12
- Dimensions with evidence
- 4/4
How to read the benchmark bars
Each bar compares models only within the same benchmark column. Bar lengths are relative to the models shown here; they are not Category Scores and cannot be compared across benchmark columns.
| Rank | Model | LMSpeed score | Abstract logic | Scientific & causal reasoning | Multi-step constraints | Evidence integration & verification | Status | Evidence | Updated | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ARC-AGI-2undefined models | GPQAundefined models | HLEundefined models | CritPtundefined models | HLE w/o toolsundefined models | AA-GPQA Diamondundefined models | AA-HLEundefined models | MMLU-Proundefined models | AA-LCRundefined models | LongBench v2undefined models | MRCR 1Mundefined models | AI-Needleundefined models | ||||||
| Formally ranked models64 | |||||||||||||||||
| 1 | GPT-5.6 SolOpenAI | 61.2±8.7 | 92.5 | 93.1% | 47.3% | 32.3 | — | — | — | — | 77.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 2 | DeepSeek V4 Pro 0813DeepSeek | 61±8.3 | — | — | — | 18.0 | — | 92.8 | 41.0 | 87.5% | 75.3 | — | 83.5 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 3 | Claude Opus 4.5Anthropic | 60.7±8.1 | — | 86.6% | 30.1% | 0.3 | — | — | — | 89.5% | 67.3 | 64.4 | — | 74.0 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 4 | Claude Opus 5Anthropic | 60.3±8.7 | 90.4 | 91.9% | 51.3% | 29.1 | 56.3 | — | — | — | 75.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 5 | Qwen3.7 MaxQwen | 59.6±8.7 | — | 92.3% | 40.5% | 13.4 | — | — | — | 89.6% | 74.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 6 | DeepSeek V4 Flash 0731DeepSeek | 59.4±8.3 | — | — | — | 16.6 | — | 90.8 | 38.6 | 86.2% | 74.3 | — | 78.7 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 7 | GPT-5.5OpenAI | 59.2±8.7 | 85.0 | 92.6% | 42.4% | 27.1 | 41.4 | — | — | — | 79.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 8 | GPT-5.6 TerraOpenAI | 58.7±8.7 | 83.9 | 89.6% | 38.5% | 30.0 | — | — | — | — | 79.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 9 | Gemini 3.1 ProGoogle | 58.2±8.7 | 77.1 | — | — | 17.7 | 45.4 | 94.1 | 47.0 | — | 79.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 10 | Qwen3.7 PlusQwen | 57.4±8.7 | — | 90.0% | 35.6% | 9.1 | — | — | — | 88.5% | 69.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 11 | Claude Opus 4.7 MaxAnthropic | 56.7±8.7 | 75.8 | — | — | 12.0 | 46.9 | 91.4 | 42.3 | — | 75.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 12 | Qwen3.6 PlusQwen | 56.6±8.1 | — | 88.2% | 27.8% | 2.9 | — | — | — | 88.5% | 72.3 | 62.0 | — | 68.3 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 13 | Claude Opus 4.8Anthropic | 56.4±8.7 | 72.1 | 92.0% | 48.7% | 20.9 | 49.8 | — | — | — | 73.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 14 | GPT-5.4OpenAI | 56.2±8.7 | 74.0 | 87.1% | 30.8% | 23.4 | 39.8 | — | — | — | 77.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 15 | Kimi K2.5MoonshotAI | 55.5±8.3 | — | 87.9% | 30.7% | 3.1 | — | — | — | 87.1% | 73.0 | 61.0 | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 16 | GPT-5.1 CodexOpenAI | 55.3±8.7 | — | 86.0% | 25.7% | 5.7 | — | — | — | 86.0% | 69.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 17 | DeepSeek V4 ProDeepSeek | 54.8±9.0 | — | 92.8% | 41.0% | — | — | — | — | 82.9% | — | — | 44.7 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 18 | GPT-5.6 LunaOpenAI | 54.4±8.7 | 59.5 | 85.9% | 25.8% | 20.6 | — | — | — | — | 78.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 19 | GLM-4.7Z.ai | 54.4±8.7 | — | 85.9% | 27.4% | 1.7 | — | — | — | 85.6% | 68.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 20 | Qwen3.6 27BQwen | 54.4±8.7 | — | 82.9% | 15.1% | 1.1 | — | — | — | 86.2% | 73.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 21 | Grok 4.5SpaceXAI | 54.4±8.7 | 52.6 | 93.1% | 42.7% | 15.4 | — | — | — | — | 74.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 22 | O3OpenAI | 54.3±8.7 | — | 82.7% | 20.1% | 1.1 | — | — | — | 85.3% | 73.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 23 | Gemini 2.5 ProGoogle | 54.2±8.7 | — | 84.4% | 22.5% | 2.6 | — | — | — | 86.2% | 66.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 24 | Gemini 3 ProGoogle | 54.1±6.4 | 31.1 | 90.8% | 39.7% | 9.1 | — | — | — | 89.8% | 73.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 25 | GPT-5.1OpenAI | 54±8.7 | — | 64.3% | 5.3% | 4.9 | — | — | — | 87.0% | 76.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 26 | DeepSeek V4 FlashDeepSeek | 53.7±9.0 | — | 90.8% | 38.6% | — | — | — | — | 83.0% | — | — | 37.5 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 27 | Qwen3.5-27BQwen | 53.6±8.3 | — | 85.8% | 23.9% | 0.9 | — | — | — | 86.1% | 72.3 | 60.6 | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 28 | GPT-5.2OpenAI | 53.5±6.4 | 52.9 | 86.4% | 26.7% | 11.6 | — | — | — | 81.4% | 79.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 29 | Claude Opus 4.6Anthropic | 53.4±8.7 | — | 89.6% | 39.9% | 2.8 | 40.0 | — | — | 82.0% | 62.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 30 | Gemini 3.5 FlashGoogle | 53.4±8.4 | 72.1 | 92.1% | 41.3% | 13.1 | — | — | — | — | 69.3 | — | 26.6 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 31 | Qwen3.5-122B-A10BQwen | 53.1±8.3 | — | 85.7% | 25.2% | 0.6 | — | — | — | 86.7% | 70.3 | 60.2 | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 32 | GPT-5OpenAI | 52.7±8.7 | — | 68.6% | 6.6% | 5.7 | — | — | — | 82.0% | 76.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 33 | Qwen3.6 35B A3BQwen | 52.4±8.7 | — | 81.7% | 13.9% | 0.3 | — | — | — | 85.2% | 66.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 34 | Qwen3.5 | 52.3±8.1 | — | 45.6% | 2.6% | 1.7 | — | — | — | 87.8% | 72.7 | 63.2 | — | 68.7 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 35 | Gemini 3 FlashGoogle | 52.2±8.7 | — | 81.2% | 15.0% | 1.4 | — | — | — | 88.2% | 53.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 36 | Qwen3.5-35B-A3BQwen | 51.2±8.3 | — | 84.5% | 21.0% | 0.9 | — | — | — | 85.3% | 68.3 | 59.0 | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 37 | DeepSeek V3.2DeepSeek | 51±8.7 | — | 87.1% | 28.7% | 0.9 | — | — | — | 86.3% | 42.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 38 | Inkling SmallThinking Machines | 50.9±8.7 | 40.1 | 89.5% | 33.3% | 8.3 | 31.6 | — | — | — | 69.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 39 | O1OpenAI | 50.7±8.7 | — | 76.5% | 7.0% | 0.3 | — | — | — | 84.8% | 63.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 40 | DeepSeek R1DeepSeek | 50.2±8.7 | — | 70.8% | 8.5% | 1.4 | — | — | — | 84.9% | 56.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 41 | Claude Sonnet 4.6Anthropic | 49.8±8.7 | — | 79.7% | 11.2% | 0.9 | — | — | — | 79.2% | 62.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 42 | gpt-oss-120bOpenAI | 49.2±8.7 | — | 78.2% | 19.6% | 1.1 | — | — | — | 80.8% | 51.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 43 | Gemini 2.5 FlashGoogle | 48.9±8.7 | — | 76.6% | 8.7% | 1.4 | — | — | — | 83.6% | 48.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 44 | GLM-5Z.ai | 48.9±8.1 | — | 66.6% | 7.6% | 2.0 | — | — | — | 85.7% | 70.7 | 60.8 | — | 63.3 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 45 | Qwen3 MaxQwen | 48.9±8.7 | — | 76.4% | 11.9% | 0.0 | — | — | — | 84.1% | 48.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 46 | DeepSeek V3.1DeepSeek | 48.7±8.7 | — | 77.9% | 14.3% | 0.0 | — | — | — | 83.3% | 46.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 47 | Kimi K2MoonshotAI | 48.6±8.7 | — | 76.6% | 7.4% | 0.0 | — | — | — | 82.4% | 53.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 48 | GPT-4.1OpenAI | 48.5±8.7 | — | 66.6% | 4.2% | 0.0 | — | — | — | 80.6% | 64.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 49 | MiMo-V2-Flash | 48.4±8.7 | — | 84.6% | 22.8% | 0.0 | — | — | — | 84.3% | 35.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 50 | Claude Sonnet 4Anthropic | 47.4±8.7 | — | 68.3% | 4.3% | 1.1 | — | — | — | 83.7% | 45.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 51 | GLM-4.5 AirZ.ai | 47±8.7 | — | 73.3% | 7.0% | 0.0 | — | — | — | 81.5% | 45.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 52 | Llama 4 MaverickMeta | 46.8±8.7 | — | 67.1% | 4.9% | 0.0 | — | — | — | 80.9% | 50.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 53 | Command ACohere | 45.5±8.7 | — | 76.1% | 12.0% | 0.3 | — | — | — | 71.2% | 48.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 54 | GPT-4.1 MiniOpenAI | 45.4±8.7 | — | 66.4% | 5.0% | 0.0 | — | — | — | 78.1% | 45.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 55 | Mistral Large 3 | 44.5±8.7 | — | 68.0% | 4.2% | 0.0 | — | — | — | 80.7% | 34.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 56 | DeepSeek V3 | 44.4±8.7 | — | 65.5% | 4.7% | 0.0 | — | — | — | 81.9% | 31.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 57 | gpt-oss-20bOpenAI | 44.2±8.7 | — | 68.8% | 11.0% | 1.4 | — | — | — | 74.8% | 33.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 58 | GLM-4.6Z.ai | 42.8±8.7 | — | 63.2% | 5.5% | 0.0 | — | — | — | 78.4% | 28.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 59 | Mistral Medium 3Mistral | 42.3±8.7 | — | 57.8% | 4.1% | 0.0 | — | — | — | 76.0% | 31.7 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 60 | Llama 4 ScoutMeta | 41.8±8.7 | — | 58.7% | 3.8% | 0.0 | — | — | — | 75.2% | 30.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 61 | GPT-4oOpenAI | 39.9±8.7 | — | 51.1% | 2.7% | 0.0 | — | — | — | 77.3% | 0.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 62 | DeepSeek R1 Distill QwenDeepSeek | 39.1±8.9 | — | 48.4% | 4.1% | — | — | — | — | 74.0% | 8.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 63 | Phi 4Microsoft | 39.1±8.7 | — | 57.5% | 3.8% | 0.0 | — | — | — | 71.4% | 0.0 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 64 | GPT-4.1 NanoOpenAI | 37.6±8.7 | — | 51.2% | 3.8% | 0.0 | — | — | — | 65.7% | 19.3 | — | — | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| Estimated models — unranked36 | |||||||||||||||||
| — | Qwen3.8 MaxQwen | 64.5±10.1 | — | 92.7% | 43.0% | 20.0 | 43.6 | — | — | — | 74.3 | 66.3 | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Muse Spark 1.2Meta | 61.3±10.8 | — | 90.4% | 45.5% | 17.7 | — | — | — | — | 83.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Fable 5Anthropic | 60.6±10.8 | — | 92.6% | 55.5% | 28.6 | — | — | — | — | 76.7 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3 Flash PreviewGoogle | 60.2±11.1 | — | 89.8% | 36.6% | — | — | — | — | 89.0% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.6 FlashGoogle | 59.7±10.8 | — | 92.8% | 40.8% | 10.6 | — | — | — | — | 79.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.3 CodexOpenAI | 59.7±10.8 | — | 91.5% | 42.5% | 16.9 | — | — | — | — | 78.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Muse Spark 1.1Meta | 59.5±10.2 | — | 89.8% | 46.2% | 15.1 | 52.2 | — | — | — | 81.3 | — | 54.1 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.7 FlashGoogle | 59.3±10.8 | — | 90.1% | 35.1% | 14.3 | — | — | — | — | 80.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Grok 4.6SpaceXAI | 59.3±10.8 | — | 93.5% | 44.1% | 17.1 | — | — | — | — | 75.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K3MoonshotAI | 59.3±10.8 | — | 84.2% | 25.0% | 23.4 | 43.5 | — | — | — | 82.7 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiniMax M3MiniMax | 59.2±10.8 | — | 92.9% | 39.0% | 3.7 | — | — | — | — | 80.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4 ProOpenAI | 59.1±11.3 | 83.3 | — | 58.7% | 30.0 | 42.7 | — | — | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.2 CodexOpenAI | 58.7±10.8 | — | 89.9% | 35.7% | 8.7 | — | — | — | — | 79.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.7 CodeMoonshotAI | 57.5±10.8 | — | 89.6% | 35.0% | 10.0 | — | — | — | — | 75.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiniMax M2.1MiniMax | 56.8±11.1 | — | 83.0% | 23.2% | — | — | — | — | 87.5% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Hy3Tencent | 56.7±10.8 | — | 89.7% | 33.5% | 4.9 | — | — | — | — | 74.7 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5 CodexOpenAI | 56.7±11.1 | — | 83.7% | 27.8% | — | — | — | — | 86.5% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Sonnet 5Anthropic | 56.3±10.8 | — | 80.0% | 19.0% | 16.9 | 43.2 | — | — | — | 77.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Muse Glimmer 30BMeta | 56.1±10.8 | — | — | — | 2.6 | — | 83.5 | 22.0 | — | 80.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.7Anthropic | 55.9±10.8 | — | 88.5% | 33.3% | 5.1 | — | — | — | — | 72.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4 MiniOpenAI | 55.6±10.8 | — | 87.5% | 28.1% | 10.0 | 28.2 | — | — | — | 73.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.6 Max PreviewQwen | 55.5±10.8 | — | 88.8% | 30.8% | 3.7 | — | — | — | — | 72.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2 ThinkingMoonshotAI | 55.5±11.1 | — | 83.8% | 23.8% | — | — | — | — | 84.8% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | InklingThinking Machines | 55.4±10.8 | — | 87.2% | 31.9% | 5.4 | 30.0 | — | — | — | 73.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 2.5 Pro Preview 06-05Google | 55.4±11.1 | — | 83.6% | 18.0% | — | — | — | — | 85.8% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.6MoonshotAI | 55.3±10.8 | — | 78.8% | 19.6% | 8.0 | — | — | — | — | 76.7 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3 Max ThinkingQwen | 55.1±11.1 | — | 86.1% | 28.0% | — | — | — | — | 82.4% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiniMax M2.7MiniMax | 54.6±10.8 | — | 87.4% | 29.6% | 0.6 | — | — | — | — | 75.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5.2Z.ai | 54.5±10.8 | — | 68.6% | 9.8% | 20.9 | 40.5 | — | — | — | 76.7 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.1 Codex MaxOpenAI | 54.3±10.8 | — | — | — | 5.7 | — | 86.0 | 25.7 | — | 69.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Grok 4.3SpaceXAI | 54.3±10.8 | — | 89.0% | 30.0% | 8.0 | — | — | — | — | 64.3 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiMo-V2.5-ProXiaomi | 54.2±10.8 | — | 76.2% | 14.8% | 4.0 | 34.0 | — | — | — | 77.7 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek V3.1 TerminusDeepSeek | 54.1±11.1 | — | 79.2% | 16.4% | — | — | — | — | 85.1% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek V3.2 ExpDeepSeek | 54±11.1 | — | 79.7% | 14.9% | — | — | — | — | 85.0% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5.1Z.ai | 53.8±10.8 | — | 83.9% | 27.9% | 4.6 | — | — | — | — | 68.0 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3 235B A22B Instruct 2507Qwen | 53.6±11.1 | — | 79.0% | 15.9% | — | — | — | — | 84.3% | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
What this leaderboard measures
Which AI model is better suited to complex reasoning?
This leaderboard covers logic, causality, multi-step constraints, and evidence verification. It measures reasoning performance, not a model's total ability across every task.
Four capability dimensions
The four dimensions come from the category blueprint. Available data may cover only some of them. A dimension without evidence is not presented as a verified capability.
Abstract logic
1 benchmark columns currently provide evidence for this dimension.
- ARC-AGI-2
Scientific & causal reasoning
6 benchmark columns currently provide evidence for this dimension.
- GPQA
- HLE
- CritPt
- HLE w/o tools
- AA-GPQA Diamond
- AA-HLE
Multi-step constraints
1 benchmark columns currently provide evidence for this dimension.
- MMLU-Pro
Evidence integration & verification
4 benchmark columns currently provide evidence for this dimension.
- AA-LCR
- LongBench v2
- MRCR 1M
- AI-Needle
Reasoning tasks this page can help with
- Complex analysis that requires keeping several conditions and intermediate conclusions aligned.
- Scientific questions that require causal reasoning, hypotheses, and evidence.
- Conclusion checks that look for gaps and test whether evidence supports a claim.
How to choose a model with this leaderboard
- Step 1
Check the rating status first
Only Rated models receive a rank. Estimated and Provisional models do not have a formal position.
- Step 2
Review uncertainty and evidence
When scores are close, do not rely on rank alone. Check uncertainty, dimension coverage, and benchmark count.
- Step 3
Test the real task last
A leaderboard cannot replace your own test. Check quality, speed, price, context, and provider limits together.
Do not rely on one high score. Review the uncertainty range, task type, answer stability, and cost of verification.
Rating status guide
Rated
Rated means the evidence and overlap rules are met. The model can receive a formal rank.
Estimated
Estimated means there is useful evidence, but it is not enough for a formal rank.
Provisional
Provisional means evidence is limited or dimension and benchmark-family coverage is below the estimated threshold. Use the result only as an early signal.
Benchmarks and evidence sources
Evidence source names and benchmark groups come from the currently available score data. One source may contribute several benchmarks.
Artificial Analysis
GPQA, HLE, and MMLU-Pro
BenchLM
AA-GPQA Diamond, AA-HLE, AA-LCR, AI-Needle, ARC-AGI-2, CritPt, GPQA, HLE, HLE w/o tools, LongBench v2, MMLU-Pro, and MRCR 1M
How the reasoning model ranking is built
LMSpeed combines eligible third-party benchmarks inside four fixed capability dimensions. Rated models meet the evidence and overlap requirements for a formal rank; Estimated and Provisional models remain visible without receiving a rank.
Read the Category Score methodologyLeaderboard limits
Category Scores use the third-party benchmarks currently included by LMSpeed. Tests can use different data, prompts, and scoring rules. The result is not permanent and cannot represent every real task. Test important choices with your own data and workflow.
Frequently asked questions
Which visible model has the highest formal rank now?
Among the currently visible models, GPT-5.6 Sol has the highest formal position at global rank 1. Its Category Score is 61.2. 64 visible models meet the formal ranking rules. This result applies only to the run date and methodology version shown on the page.
Can I compare scores across different categories?
No. Each category uses different capability dimensions and evidence. A Category Score is comparable only inside the same leaderboard. Review the matching category for each task.
Are Estimated and Provisional models still useful?
They can help you find candidates, but their evidence is not complete enough for a formal rank. Review coverage and uncertainty, then test the model on a real task.
How often does the leaderboard update?
The leaderboard updates after a new completed score run is published. The current run date and methodology version appear above. LMSpeed does not promise a fixed daily or weekly schedule.
Is the number one model always best for me?
No. Your result also depends on speed, price, context length, tool support, region, and provider limits. Use the leaderboard to narrow the field, then run your own test.
Is a reasoning score the same as a math score?
No. Math is a separate category. The reasoning leaderboard also covers abstract logic, scientific causality, multi-step constraints, and evidence verification. A strong math score does not prove better results on every reasoning task.