LiveBench 2026 Coding Leaderboard
LiveBench 2026 Coding measures whether a model writes correct code for programming problems. 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 Coding
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 | Claude Fable 5 · proprietary | 98.3 | API only |
| 2 | Claude Fable 5.1 · proprietary | 97.7 | API only |
| 3 | Claude Opus 5.5 · proprietary | 96.3 | API only |
| 4 | GPT 6.1 Sol · proprietary | 96.0 | API only |
| 5 | Claude Sonnet 5.5 · proprietary | 94.7 | API only |
| 6 | Grok 4.5 · proprietary | 94.3 | API only |
| 7 | GPT 6 Astra · proprietary | 93.7 | API only |
| 8 | GPT 6 Sol · proprietary | 93.1 | API only |
| 9 | Claude Opus 5 · proprietary | 90.8 | API only |
| 10 | Muse Spark 1.3 · proprietary | 90.2 | API only |
| 11 | Gemini 4 Argon · proprietary | 89.7 | API only |
| 12 | GPT 5.6 Sol · proprietary | 89.7 | API only |
| 13 | Grok 4.6 · proprietary | 89.1 | API only |
| 14 | Qwen3.8 Flash · proprietary | 87.4 | API only |
| 15 | Qwen3.8 27B · 27.8B | 86.8 | ~17 GB |
| 16 | GPT 5.6 Terra · proprietary | 86.2 | API only |
| 17 | Muse Wonder 1.0 · proprietary | 86.2 | API only |
| 18 | Step 5 · proprietary | 85.6 | API only |
| 19 | DeepSeek V4.1 Flash · 763.2B | 85.4 | ~458 GB |
| 20 | GPT 5.5 · proprietary | 85.1 | API only |
| 21 | Claude Haiku 5.5 · proprietary | 84.5 | API only |
| 22 | GPT 6 Luna · proprietary | 84.5 | API only |
| 23 | Grok 4.7 · proprietary | 84.5 | API only |
| 24 | Beam 1.0 · proprietary | 83.9 | API only |
| 25 | GLM 5.3 · 753.3B | 82.8 | ~456 GB |
| 26 | Claude Sonnet 5 · proprietary | 82.2 | API only |
| 27 | Gemini 3.8 Flash · proprietary | 82.2 | API only |
| 28 | MiMo V2.6 Pro RL · 1020B | 81.5 | ~613 GB |
| 29 | GLM 5.3 Flash · 321.3B | 80.6 | ~195 GB |
| 30 | Muse Spark 1.2 · proprietary | 80.6 | API only |
| 31 | Gemini 3.7 Flash · proprietary | 80.5 | API only |
| 32 | Claude Opus 4.8 · proprietary | 77.3 | API only |
| 33 | GPT 5.6 Luna · proprietary | 73.6 | API only |
| 34 | Muse Spark Open 816B A42B · 822.0B | 73.6 | ~543 GB |
| 35 | MiniMax M3 · 427.0B | 70.9 | ~257 GB |
| 36 | Gemini 3.1 Pro · proprietary | 66.7 | API only |
| 37 | Claude Sonnet 4.6 · proprietary | 66.6 | API only |
| 38 | NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124.5B | 65.5 | ~82 GB |
| 39 | Inkling · 952.4B | 64.0 | ~572 GB |
| 40 | Qwen3.8 Max · proprietary | 60.3 | API only |
| 41 | Smaug Flash · 304.2B | 60.1 | ~609 GB |
| 42 | Mistral Large 4 · proprietary | 60.0 | API only |
| 43 | Kimi K3 · 2779.9B | 59.3 | ~1674 GB |
| 44 | Gemma 4 31B IT · 31.3B | 49.0 | ~20 GB |
| 45 | Claude Haiku 4.5 · proprietary | 48.1 | API only |
| 46 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 43.6 | ~370 GB |
| 47 | DeepSeek V4 Flash Vision Exp · 304.6B | 37.6 | ~183 GB |
| 48 | NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 · 31.6B | 22.6 | ~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 86.8 — on the efficiency frontier (best score at its size or smaller).
LiveBench 2026 Coding: frequently asked questions
- What is the best open LLM on LiveBench 2026 Coding?
- Qwen3.8 27B is the top open model on LiveBench 2026 Coding, scoring 86.8. Among all models tested — including proprietary ones — it ranks #15. The top model overall is Claude Fable 5 (Anthropic) at 98.3.
- What's the best LiveBench 2026 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 (~17 GB), scoring 86.8 on LiveBench 2026 Coding.
- Can open models match proprietary models on LiveBench 2026 Coding?
- Not quite on LiveBench 2026 Coding: the strongest proprietary model (Claude Fable 5) scores 98.3, ahead of the best open model (Qwen3.8 27B) at 86.8 — 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.