llmrun
Reasoning

LiveBench Reasoning Leaderboard

LiveBench Reasoning measures logical, multi-step reasoning using contamination-free questions that are refreshed regularly, so models cannot have trained on the test set.

Source: livebench8 open models ranked+29 proprietaryData through Jun 2026

Open modelsAll models

All models ranked on LiveBench Reasoning

Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.

#ModelScore
1GPT 5.6 Sol Max · proprietary
91.7
2Kimi K3 · proprietary
90.7
3GPT 5.6 Terra Max · proprietary
90.6
4GPT 5.6 Sol · proprietary
90.2
5Claude Opus 4.8 · proprietary
89.7
6Claude Fable 5 Max · proprietary
89.7
7GPT 5.5 · proprietary
89.7
8Claude Sonnet 5 · proprietary
88.7
9Claude Opus 4.6 · proprietary
88.7
10GPT 5.4 · proprietary
88.1
11Muse Spark 1.1 · proprietary
87.7
12Claude Fable 5 · proprietary
87.7
13Claude Opus 4.7 · proprietary
87.2
14Grok 4.5 · proprietary
87.2
15GPT 5.6 Luna Max · proprietary
85.6
16GPT 5.6 Terra · proprietary
84.9
17Claude Sonnet 4.6 · proprietary
84.8
18GPT 5.6 Luna · proprietary
84.7
19Gemini 3.1 Pro · proprietary
84.0
20Qwen3.7 Max · proprietary
83.3
21GPT 5.2 · proprietary
83.2
22Kimi K2.7 Code · 1058.6B
82.8
23DeepSeek V4 Pro · 861.6B
82.7
24Gemini 3.5 Flash · proprietary
82.0
25GPT 5.4 Nano · proprietary
81.1
26Claude Opus 4.5 · proprietary
80.1
27Kimi K2.6 · 1058.6B
79.4
28GLM 5.2 · 753.3B
78.6
29Inkling · 952.4B
78.3
30GPT 5.2 Codex · proprietary
77.7
31Grok Build 0.1 · proprietary
76.4
32Qwen3.6 Plus · proprietary
75.8
33MiniMax M3 · 427.0B
74.5
34GPT 5.4 Mini · proprietary
71.3
35Grok 4.3 · proprietary
70.8
36DeepSeek V4 Flash · 158.1B
70.6
37Qwen3.6 27B · 27.8B
70.3

Score vs model size

Which models give the most quality for their size — the ones worth running locally.

100B1Tmodel size (log scale) →82.870.3Kimi K2.6 · 1.1T · 79.4Inkling · 952B · 78.3Qwen3.6 27B · 28B · 70.3Qwen3.6 27BDeepSeek V4 Flash · 158B · 70.6DeepSeek V4 FlashMiniMax M3 · 427B · 74.5MiniMax M3GLM 5.2 · 753B · 78.6GLM 5.2DeepSeek V4 Pro · 862B · 82.7DeepSeek V4 ProKimi K2.7 Code · 1.1T · 82.8Kimi K2.7 Code
Each dot is a model. Up = higher score, left = smaller (easier to run locally). The dashed line marks the efficiency frontier — the best score you can get at each size or smaller.
  • Qwen3.6 27B, 28B, score 70.3 — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Flash, 158B, score 70.6 — on the efficiency frontier (best score at its size or smaller).
  • MiniMax M3, 427B, score 74.5 — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.2, 753B, score 78.6 — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Pro, 862B, score 82.7 — on the efficiency frontier (best score at its size or smaller).
  • Kimi K2.7 Code, 1.1T, score 82.8 — on the efficiency frontier (best score at its size or smaller).

LiveBench Reasoning: frequently asked questions

What is the best open LLM on LiveBench Reasoning?
Kimi K2.7 Code is the top open model on LiveBench Reasoning, scoring 82.8. Among all models tested — including proprietary ones — it ranks #22. The top model overall is GPT 5.6 Sol Max (OpenAI) at 91.7.
What's the best LiveBench Reasoning model you can run on a 24 GB GPU?
Qwen3.6 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 15 GB), scoring 70.3 on LiveBench Reasoning.
Can open models match proprietary models on LiveBench Reasoning?
Not quite on LiveBench Reasoning: the strongest proprietary model (GPT 5.6 Sol Max) scores 91.7, ahead of the best open model (Kimi K2.7 Code) at 82.8 — 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.