llmrun
Coding

LiveBench Coding Leaderboard

LiveBench Coding evaluates code generation and completion on fresh, contamination-free programming tasks that are updated regularly.

Source: livebench16 open models ranked+38 proprietaryData through Jun 2026

Open modelsAll models

All models ranked on LiveBench Coding

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

#ModelScore
1Claude Fable 5.1 Max · proprietary
86.4
2Claude Fable 5 Max · proprietary
86.0
3GPT 5.6 Sol Max · proprietary
83.9
4GPT 5.2 Codex · proprietary
83.6
5GPT 5.6 Luna Max · proprietary
82.9
6Smaug Agentic · proprietary
82.5
7GPT 5.5 · proprietary
82.2
8Claude Opus 4.7 · proprietary
82.1
9Claude Opus 4.8 Max · proprietary
81.8
10Claude Opus 5 Max · proprietary
81.5
11Kimi K3 · 2779.9B
81.5
12Muse Spark 1.3 · proprietary
81.1
13Claude Sonnet 5 · proprietary
80.7
14GPT 6 Astra Max · proprietary
80.4
15Claude Opus 4.5 · proprietary
79.7
16GLM 5.2 · 753.3B
79.7
17Claude Sonnet 4.6 · proprietary
79.3
18GLM 5.3 · 753.3B
79.0
19GLM 5.3 Flash · 321.3B
79.0
20Gemini 3.7 Flash · proprietary
78.9
21Kimi K2.6 · 1026.9B
78.6
22GPT 5.6 Terra Max · proprietary
78.3
23Claude Opus 4.6 · proprietary
78.2
24Gemini 3.5 Flash · proprietary
78.2
25Qwen3.6 Plus · proprietary
78.2
26Gemini 3.6 Flash · proprietary
77.9
27GPT 5.4 · proprietary
77.5
28Muse Spark 1.2 · proprietary
77.5
29DeepSeek V4 Pro 0813 · 1650.5B
77.2
30Muse Spark 1.1 · proprietary
77.2
31Grok 4.6 · proprietary
76.8
32Gemini 3.1 Pro · proprietary
76.5
33Gemini 3.5 Flash Lite · proprietary
76.1
34GPT 5.2 · proprietary
76.1
35Ox Alpha Max · proprietary
75.8
36Qwen3.8 27B · 27.8B
75.7
37DeepSeek V4 Flash 0731 · 304.2B
75.0
38Qwen3.7 Max · proprietary
74.2
39Kimi K2.7 Code · 1026.9B
74.0
40Qwen3.8 Max · proprietary
72.9
41Qwen3.8 Flash Next · 180.0B
72.5
42Gemini 3.8 Flash · proprietary
72.5
43Qwen3.6 27B · 27.8B
71.8
44GPT 5.4 Mini · proprietary
71.6
45Inkling · 952.4B
71.0
46GPT 5.4 Nano · proprietary
70.8
47NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B
70.7
48DeepSeek V4 Pro · 1598.8B
70.0
49Grok 4.3 · proprietary
69.9
50DeepSeek V4 Flash · 290.9B
69.2
51Grok 4.5 · proprietary
68.6
52DeepSeek v4 Flash Vision · proprietary
68.2
53MiniMax M3 · 427.0B
68.2
54Grok Build 0.1 · proprietary
65.4

Score vs model size

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

100B1Tmodel size (log scale) →81.568.2GLM 5.3 · 753B · 79.0Kimi K2.6 · 1T · 78.6DeepSeek V4 Pro 0813 · 1.7T · 77.2DeepSeek V4 Flash 0731 · 304B · 75.0Kimi K2.7 Code · 1T · 74.0Qwen3.8 Flash Next · 180B · 72.5Qwen3.6 27B · 28B · 71.8Inkling · 952B · 71.0NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 561B · 70.7DeepSeek V4 Pro · 1.6T · 70.0DeepSeek V4 Flash · 291B · 69.2MiniMax M3 · 427B · 68.2Qwen3.8 27B · 28B · 75.7Qwen3.8 27BGLM 5.3 Flash · 321B · 79.0GLM 5.3 FlashGLM 5.2 · 753B · 79.7GLM 5.2Kimi K3 · 2.8T · 81.5Kimi K3
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.
  • Qwen3.8 27B, 28B, score 75.7 — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.3 Flash, 321B, score 79.0 — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.2, 753B, score 79.7 — on the efficiency frontier (best score at its size or smaller).
  • Kimi K3, 2.8T, score 81.5 — on the efficiency frontier (best score at its size or smaller).

LiveBench Coding: frequently asked questions

What is the best open LLM on LiveBench Coding?
Kimi K3 is the top open model on LiveBench Coding, scoring 81.5. Among all models tested — including proprietary ones — it ranks #10. The top model overall is Claude Fable 5.1 Max (Anthropic) at 86.4.
What's the best LiveBench Coding 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 75.7 on LiveBench Coding.
Can open models match proprietary models on LiveBench Coding?
Not quite on LiveBench Coding: the strongest proprietary model (Claude Fable 5.1 Max) scores 86.4, ahead of the best open model (Kimi K3) at 81.5 — 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.