What is the best open LLM on LiveBench Reasoning?

Kimi K3 is the top open model on LiveBench Reasoning, scoring 90.7. Among all models tested — including proprietary ones — it ranks #5. The top model overall is GPT 6 Astra Max (OpenAI) at 92.7.

What's the best LiveBench 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 (about 15 GB), scoring 80.0 on LiveBench Reasoning.

Can open models match proprietary models on LiveBench Reasoning?

Not quite on LiveBench Reasoning: the strongest proprietary model (GPT 6 Astra Max) scores 92.7, ahead of the best open model (Kimi K3) at 90.7 — but you can run the open one yourself.

Reasoning

LiveBench Reasoning Leaderboard

Name: LiveBench Reasoning — open LLM scores
Creator: livebench

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: livebench16 open models ranked+38 proprietaryData through Jun 2026

Open models All 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.

#	Model	Score
1	GPT 6 Astra Max · proprietary	92.7
2	Claude Fable 5.1 Max · proprietary	91.7
3	GPT 5.6 Sol Max · proprietary	91.7
4	Claude Opus 5 Max · proprietary	91.2
5	Kimi K3 · 2779.9B	90.7
6	GPT 5.6 Terra Max · proprietary	90.6
7	Grok 4.6 · proprietary	90.5
8	Smaug Agentic · proprietary	90.3
9	Muse Spark 1.2 · proprietary	90.0
10	Claude Fable 5 Max · proprietary	89.7
11	GPT 5.5 · proprietary	89.7
12	Muse Spark 1.3 · proprietary	89.7
13	Gemini 3.8 Flash · proprietary	89.3
14	Claude Opus 4.8 Max · proprietary	89.2
15	Claude Sonnet 5 · proprietary	88.7
16	Claude Opus 4.6 · proprietary	88.7
17	Qwen3.8 Max · proprietary	88.2
18	GPT 5.4 · proprietary	88.1
19	Gemini 3.7 Flash · proprietary	87.8
20	Muse Spark 1.1 · proprietary	87.7
21	Qwen3.8 Flash Next · 180.0B	87.4
22	Claude Opus 4.7 · proprietary	87.2
23	Grok 4.5 · proprietary	87.2
24	DeepSeek V4 Flash 0731 · 304.2B	86.6
25	DeepSeek V4 Pro 0813 · 1650.5B	85.8
26	GLM 5.3 · 753.3B	85.8
27	GPT 5.6 Luna Max · proprietary	85.6
28	DeepSeek v4 Flash Vision · proprietary	85.4
29	Gemini 3.6 Flash · proprietary	85.2
30	Claude Sonnet 4.6 · proprietary	84.8
31	Gemini 3.1 Pro · proprietary	84.0
32	Qwen3.7 Max · proprietary	83.3
33	GPT 5.2 · proprietary	83.2
34	Kimi K2.7 Code · 1026.9B	82.8
35	DeepSeek V4 Pro · 1598.8B	82.7
36	Gemini 3.5 Flash · proprietary	82.0
37	GPT 5.4 Nano · proprietary	81.1
38	Claude Opus 4.5 · proprietary	80.1
39	Qwen3.8 27B · 27.8B	80.0
40	Kimi K2.6 · 1026.9B	79.4
41	GLM 5.2 · 753.3B	78.6
42	Inkling · 952.4B	78.3
43	GPT 5.2 Codex · proprietary	77.7
44	GLM 5.3 Flash · 321.3B	77.6
45	Ox Alpha Max · proprietary	76.6
46	Grok Build 0.1 · proprietary	76.4
47	Qwen3.6 Plus · proprietary	75.8
48	NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B	74.7
49	MiniMax M3 · 427.0B	74.5
50	GPT 5.4 Mini · proprietary	71.3
51	Grok 4.3 · proprietary	70.8
52	DeepSeek V4 Flash · 290.9B	70.6
53	Qwen3.6 27B · 27.8B	70.3
54	Gemini 3.5 Flash Lite · proprietary	60.2

Score vs model size

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

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.

LiveBench Reasoning: frequently asked questions

What is the best open LLM on LiveBench Reasoning?: Kimi K3 is the top open model on LiveBench Reasoning, scoring 90.7. Among all models tested — including proprietary ones — it ranks #5. The top model overall is GPT 6 Astra Max (OpenAI) at 92.7.
What's the best LiveBench 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 (about 15 GB), scoring 80.0 on LiveBench Reasoning.
Can open models match proprietary models on LiveBench Reasoning?: Not quite on LiveBench Reasoning: the strongest proprietary model (GPT 6 Astra Max) scores 92.7, 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.