LiveBench 2.0 Research
Search · 2026-10-14
LiveBench 2.0 Research suite: Answer an open-ended brief by researching it and writing a sourced report (regulatory, legal, tax and finance, strategy, science); 22 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 Opus 5 | 88.9 |
| Muse Wonder | 88.0 |
| Claude Fable 5 | 86.9 |
| DeepSeek V4.1 Flash | 86.8 |
| Grok 4.7 | 86.4 |
| Grok 4.5 | 85.9 |
| GPT-6 Astra | 85.2 |
| Claude Opus 5.5 | 85.0 |
| Grok 4.6 | 84.5 |
| Claude Sonnet 5.5 | 83.6 |
| GPT-6.1 Sol | 83.5 |
| Gemini 4 Argon | 83.1 |
| Sonnet 5 | 82.8 |
| Claude Fable 5.1 | 82.0 |
| Claude Haiku 5.5 | 81.9 |
| Opus 4.8 | 81.8 |
| Muse Spark Open | 80.3 |
| DeepSeek V4.1 Flash 1017 | 79.5 |
| DeepSeek-V4-Flash-0731 | 79.2 |
| Gemini 3.8 Flash | 79.1 |
| Kimi K3 | 79.0 |
| Kimi K3.1 Pro | 77.9 |
| MiMo-V2.6-Pro | 76.7 |
| Gemini 3.7 Flash | 76.5 |
| Qwen3.8-Flash-Next | 74.6 |
| GLM-5.3 | 73.2 |
| Beam 1.0 501B A23B | 72.8 |
| Step 5 Preview | 72.8 |
| MiniMax M3.1 Flash | 72.7 |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 70.9 |
| Nemotron 3.5 Super VL 120B A12B BF16 | 70.9 |
| GLM-5.3 Flash | 70.1 |
| GPT-6 Sol | 69.3 |
| DeepSeek V4 Flash Vision Exp | 69.2 |
| Muse Spark 1.3 | 68.6 |
| MiniMax M3 | 68.0 |
| Muse Spark 1.2 | 67.7 |
| Gemini 3.1 Pro Preview | 65.7 |
| Sonnet 4.6 | 63.6 |
| GPT-5.5 | 61.6 |