Reasoning
LiveBench 2026 Reasoning Leaderboard
LiveBench 2026 Reasoning measures logical, multi-step problem solving. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
Source: livebench15 open models ranked+33 proprietaryData through Sep 2026
All models ranked on LiveBench 2026 Reasoning
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | Gemini 4 Argon · proprietary | 95.1 | API only |
| 2 | GPT 6 Astra · proprietary | 92.7 | API only |
| 3 | GPT 6.1 Sol · proprietary | 92.6 | API only |
| 4 | Claude Opus 5.5 · proprietary | 92.2 | API only |
| 5 | Claude Fable 5.1 · proprietary | 91.7 | API only |
| 6 | GPT 5.6 Sol · proprietary | 91.7 | API only |
| 7 | Claude Sonnet 5.5 · proprietary | 91.6 | API only |
| 8 | Claude Opus 5 · proprietary | 91.2 | API only |
| 9 | Kimi K3 · 2779.9B | 90.7 | ~1674 GB |
| 10 | GPT 5.6 Terra · proprietary | 90.6 | API only |
| 11 | Muse Wonder 1.0 · proprietary | 90.6 | API only |
| 12 | Grok 4.6 · proprietary | 90.5 | API only |
| 13 | Muse Spark 1.2 · proprietary | 90.0 | API only |
| 14 | Claude Fable 5 · proprietary | 89.7 | API only |
| 15 | GPT 5.5 · proprietary | 89.7 | API only |
| 16 | Muse Spark 1.3 · proprietary | 89.7 | API only |
| 17 | Gemini 3.8 Flash · proprietary | 89.3 | API only |
| 18 | Claude Opus 4.8 · proprietary | 89.2 | API only |
| 19 | Claude Sonnet 5 · proprietary | 88.7 | API only |
| 20 | GPT 6 Sol · proprietary | 88.7 | API only |
| 21 | Qwen3.8 Max · proprietary | 88.2 | API only |
| 22 | Gemini 3.7 Flash · proprietary | 87.8 | API only |
| 23 | Qwen3.8 Flash · proprietary | 87.4 | API only |
| 24 | Grok 4.5 · proprietary | 87.2 | API only |
| 25 | DeepSeek V4.1 Flash · 763.2B | 86.7 | ~458 GB |
| 26 | Claude Sonnet 4.6 · proprietary | 86.4 | API only |
| 27 | Smaug Flash · 304.2B | 86.2 | ~609 GB |
| 28 | GLM 5.3 · 753.3B | 85.8 | ~456 GB |
| 29 | GPT 5.6 Luna · proprietary | 85.6 | API only |
| 30 | DeepSeek V4 Flash Vision Exp · 304.6B | 85.4 | ~183 GB |
| 31 | Claude Haiku 5.5 · proprietary | 84.6 | API only |
| 32 | Gemini 3.1 Pro · proprietary | 84.0 | API only |
| 33 | Mistral Large 4 · proprietary | 83.9 | API only |
| 34 | Step 5 · proprietary | 83.8 | API only |
| 35 | Grok 4.7 · proprietary | 82.7 | API only |
| 36 | GPT 6 Luna · proprietary | 81.8 | API only |
| 37 | Beam 1.0 · proprietary | 80.2 | API only |
| 38 | Qwen3.8 27B · 27.8B | 80.0 | ~17 GB |
| 39 | MiMo V2.6 Pro RL · 1020B | 79.6 | ~613 GB |
| 40 | NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124.5B | 78.8 | ~82 GB |
| 41 | Inkling · 952.4B | 78.3 | ~572 GB |
| 42 | GLM 5.3 Flash · 321.3B | 77.6 | ~195 GB |
| 43 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 74.7 | ~370 GB |
| 44 | MiniMax M3 · 427.0B | 74.5 | ~257 GB |
| 45 | Muse Spark Open 816B A42B · 822.0B | 73.7 | ~543 GB |
| 46 | Gemma 4 31B IT · 31.3B | 68.3 | ~20 GB |
| 47 | Claude Haiku 4.5 · proprietary | 61.7 | API only |
| 48 | NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 · 31.6B | 42.9 | ~19 GB |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Qwen3.8 27B, 28B, score 80.0 — on the efficiency frontier (best score at its size or smaller).
- Smaug Flash, 304B, score 86.2 — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4.1 Flash, 763B, score 86.7 — on the efficiency frontier (best score at its size or smaller).
- Kimi K3, 2.8T, score 90.7 — on the efficiency frontier (best score at its size or smaller).
LiveBench 2026 Reasoning: frequently asked questions
- What is the best open LLM on LiveBench 2026 Reasoning?
- Kimi K3 is the top open model on LiveBench 2026 Reasoning, scoring 90.7. Among all models tested — including proprietary ones — it ranks #9. The top model overall is Gemini 4 Argon (Google) at 95.1.
- What's the best LiveBench 2026 Reasoning 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 (~17 GB), scoring 80.0 on LiveBench 2026 Reasoning.
- Can open models match proprietary models on LiveBench 2026 Reasoning?
- Not quite on LiveBench 2026 Reasoning: the strongest proprietary model (Gemini 4 Argon) scores 95.1, ahead of the best open model (Kimi K3) at 90.7 — but you can run the open one yourself.
Scores aggregated from livebench. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.