LiveBench 2.0 Coding
Code · 2026-10-14
LiveBench 2.0 Coding suite: Fix a real bug in a large production codebase frozen at a fixed snapshot; hidden tests grade the fix and a fix that breaks existing behaviour scores 0; 29 tasks. LiveBench 2.0 (Abacus AI, released October 14, 2026) replaced the v1 question-answer categories with seven suites, five of them agentic; each suite is one run per model at its highest supported effort, a task the run could not measure scores 0, and a model without an input a suite needs (a text-only model on Browser) scores 0 there. Not comparable with LiveBench v1.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 98.3 |
| Claude Fable 5.1 | 97.7 |
| Claude Opus 5.5 | 96.3 |
| DeepSeek V4.1 Flash 1017 | 96.0 |
| GPT-6.1 Sol | 96.0 |
| Claude Sonnet 5.5 | 94.7 |
| Grok 4.5 | 94.3 |
| GPT-6 Astra | 93.7 |
| GPT-6 Sol | 93.1 |
| Claude Haiku 5.5 | 91.4 |
| Claude Opus 5 | 90.8 |
| DeepSeek V4.1 Flash | 90.2 |
| Muse Spark 1.3 | 90.2 |
| Gemini 4 Argon | 89.7 |
| GPT-5.6 Sol | 89.7 |
| Grok 4.6 | 89.1 |
| Qwen3.8-Flash-Next | 87.4 |
| Qwen3.8 27B | 86.8 |
| GPT-5.6 Terra | 86.2 |
| Muse Wonder | 86.2 |
| Step 5 Preview | 85.6 |
| GPT-5.5 | 85.1 |
| GPT-6 Luna | 84.5 |
| Grok 4.7 | 84.5 |
| Beam 1.0 501B A23B | 83.9 |
| MiniMax M3.1 Flash | 83.9 |
| GLM-5.3 | 82.8 |
| Gemini 3.8 Flash | 82.2 |
| Kimi K3.1 Pro | 81.8 |
| MiMo-V2.6-Pro | 81.5 |
| GLM-5.3 Flash | 80.6 |
| Gemini 3.7 Flash | 80.5 |
| Muse Spark 1.2 | 78.2 |
| Opus 4.8 | 77.3 |
| DeepSeek-V4-Flash-0731 | 77.0 |
| Sonnet 5 | 74.7 |
| GPT-5.6 Luna | 73.6 |
| Muse Spark Open | 73.6 |
| MiniMax M3 | 70.9 |
| Gemini 3.1 Pro Preview | 66.7 |