LiveBench research leaderboard
- Claude Opus 5 leads
- 88.9%
- models measured
- 43
- lead over #2
- +2.0 pts
- median
- 74.6%
Leader on LiveBench research · read October 21
On October 21, 2026, 43 of our catalog's models had a LiveBench research result, and Anthropic Claude Opus 5 led with 88.9%, ahead of Anthropic Claude Fable 5 at 86.9%.
Open-ended research briefs that produce a written report, graded against a rubric. Each score is the model's mean task score for the category, 0 to 100, from a single run of the current release, shown as a share.
Role in the index
LiveBench research carries 30% of the Knowledge Index. A model missing it is scored on the other benchmarks of that index. Read the methodology
| # | Model | Result | Cost per task | Output tokens per task | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 88.9% | $16.9704 | 76,872 | 83.0 | $25.00 | |
| 2 | Claude Fable 5Anthropic | 86.9% | $17.2743 | 71,540 | 79.3 | $50.00 | |
| 3 | DeepSeek V4.1 FlashDeepSeek | 86.8% | $0.4087 | 142,711 | 62.5 | $0.78 | |
| 4 | Grok 4.7xAI | 86.4% | $4.5773 | 76,863 | 66.7 | $6.00 | |
| 5 | Grok 4.5xAI | 85.9% | $0.8179 | 15,217 | 58.6 | $6.00 | |
| 6 | GPT-6 AstraOpenAI | 85.2% | $17.6599 | 33,608 | 85.1 | $50.00 | |
| 7 | Claude Opus 5.5Anthropic | 85.0% | $8.8178 | 129,426 | 88.8 | $20.00 | |
| 8 | Grok 4.6xAI | 84.5% | $2.3255 | 28,110 | 65.7 | $6.00 | |
| 9 | Claude Sonnet 5.5Anthropic | 83.6% | $8.1856 | 239,744 | 87.3 | $10.00 | |
| 10 | GPT-6.1 SolOpenAI | 83.5% | $2.4152 | 44,333 | 82.4 | $10.00 | |
| 11 | Gemini 4 ArgonGoogle | 83.1% | $5.9556 | 144,282 | 81.0 | $10.00 | |
| 12 | Claude Sonnet 5Anthropic | 82.8% | $5.9611 | 85,447 | 57.7 | $10.00 | |
| 13 | Claude Fable 5.1Anthropic | 82.0% | $20.886 | 70,513 | 84.2 | $50.00 | |
| 14 | Claude Haiku 5.5Anthropic | 81.9% | $1.9399 | 235,636 | 70.7 | $0.50 | |
| 15 | Claude Opus 4.8Anthropic | 81.8% | $6.2353 | 64,378 | 65.0 | $25.00 | |
| 16 | Muse Spark OpenMeta | 80.3% | $0.732 | 22,757 | 47.4 | $4.25 | |
| 17 | DeepSeek V4.1 Flash 1017DeepSeek | 79.5% | $0.3754 | 124,572 | 70.3 | $0.60 | |
| 18 | Gemini 3.8 FlashGoogle | 79.1% | $2.6381 | 135,968 | 61.6 | $3.75 | |
| 19 | Kimi K3Moonshot AI | 79.0% | $2.4049 | 31,404 | 62.3 | $15.00 | |
| 20 | MiMo-V2.6-ProXiaomi | 76.7% | $0.1423 | 57,443 | 65.4 | $0.87 | |
| 21 | Gemini 3.7 FlashGoogle | 76.5% | $2.6836 | 127,694 | 58.0 | $3.75 | |
| 22 | Qwen3.8 FlashQwen | 74.6% | $0.3168 | 115,708 | Not benchmarked | $0.47 | |
| 23 | GLM 5.3Z.ai | 73.2% | $1.6346 | 78,992 | 70.6 | $4.40 | |
| 24 | Step 5 PreviewStepFun | 72.8% | $1.2289 | 72,511 | 61.1 | $2.70 | |
| 25 | Nemotron 3.5 Super VLNVIDIA | 70.9% | $0.1668 | 44,246 | Not benchmarked | $0.50 | |
| 26 | GLM 5.3 FlashZ.ai | 70.1% | $0.187 | 59,511 | 57.7 | $0.50 | |
| 27 | Qwen3.8 Max (0902)Qwen | 69.9% | $1.9547 | 105,534 | 62.4 | $6.00 | |
| 28 | GPT-6 SolOpenAI | 69.3% | $3.5051 | 42,995 | 76.0 | $10.00 | |
| 29 | DeepSeek V4 Flash Vision ExpDeepSeek | 69.2% | $0.2488 | 30,492 | Not benchmarked | $1.32 | |
| 30 | Muse Spark 1.3Meta | 68.6% | $2.6941 | 39,607 | 57.4 | $4.25 | |
| 31 | MiniMax M3MiniMax | 68.0% | $0.2516 | 34,479 | 39.2 | $1.20 | |
| 32 | Muse Spark 1.2Meta | 67.7% | $1.7832 | 28,067 | 55.7 | $4.25 | |
| 33 | Claude Sonnet 4.6Anthropic | 63.6% | $3.9018 | 35,125 | 57.3 | $15.00 | |
| 34 | GPT-5.5OpenAI | 61.6% | $17.1379 | 21,217 | 71.3 | $30.00 | |
| 35 | GPT-5.6 TerraOpenAI | 60.4% | $7.7186 | 76,949 | 63.9 | $12.00 | |
| 36 | T InklingThinking Machines | 59.9% | $1.668 | 34,077 | 31.5 | $4.05 | |
| 37 | Qwen3.8 27BQwen | 59.4% | $1.1275 | 84,200 | 74.2 | $2.55 | |
| 38 | GPT-5.6 SolOpenAI | 54.5% | $14.426 | 48,343 | 73.5 | $20.00 | |
| 39 | GPT-5.6 LunaOpenAI | 51.0% | $0.8208 | 69,492 | 58.0 | $1.20 | |
| 40 | Gemma 4 31BGoogle | 44.4% | $0.0384 | 10,393 | 39.0 | $0.40 | |
| 41 | GPT-6 LunaOpenAI | 43.5% | $0.3553 | 241,287 | 57.3 | $0.50 | |
| 42 | Claude Haiku 4.5Anthropic | 42.2% | $0.4322 | 14,045 | 35.6 | $5.00 | |
| 43 | Mistral Large 4Mistral AI | 28.1% | $2.3272 | 229,696 | 56.4 | $4.18 |
Only models in our catalog are listed, and each appears once with its best published run. 43 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out. Cost per task is what LiveBench reports the run spent on the category, divided by the number of tasks in it, in US dollars; LiveBench defines it the same way. Every other column is a figure the board publishes for the same run.
| Model | Result | Regulatory & Compliance | Legal | Tax & Finance | Strategy | Science & Evidence |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 5 | 88.9% | 89.2% | 97.2% | 79.4% | 90.6% | 91.2% |
| Anthropic Claude Fable 5 | 86.9% | 88.8% | 98.9% | 68.2% | 90.7% | 93.6% |
| DeepSeek V4.1 Flash | 86.8% | 84.8% | 91.0% | 84.6% | 85.5% | 90.6% |
| SpaceXAI Grok 4.7 | 86.4% | 83.9% | 90.8% | 81.9% | 91.6% | 88.9% |
| SpaceXAI Grok 4.5 | 85.9% | 83.5% | 89.5% | 83.5% | 90.5% | 86.3% |
| OpenAI GPT-6 Astra | 85.2% | 80.0% | 94.4% | 85.0% | 87.1% | 84.0% |
| Anthropic Claude Opus 5.5 | 85.0% | 87.2% | 99.9% | 82.0% | 87.5% | 62.4% |
| SpaceXAI Grok 4.6 | 84.5% | 85.3% | 88.5% | 79.9% | 82.9% | 86.8% |
| Anthropic Claude Sonnet 5.5 | 83.6% | 89.2% | 95.1% | 60.5% | 82.6% | 94.8% |
| OpenAI GPT-6.1 Sol | 83.5% | 77.3% | 99.8% | 81.8% | 76.0% | 86.7% |
| Google Gemini 4 Argon | 83.1% | 77.0% | 98.1% | 76.5% | 84.8% | 86.7% |
| Anthropic Claude Sonnet 5 | 82.8% | 76.4% | 97.1% | 74.2% | 87.0% | 89.2% |
| Anthropic Claude Fable 5.1 | 82.0% | 86.3% | 93.5% | 75.4% | 56.5% | 93.5% |
| Anthropic Claude Haiku 5.5 | 81.9% | 80.4% | 85.1% | 73.1% | 85.4% | 92.2% |
| Anthropic Claude Opus 4.8 | 81.8% | 86.7% | 92.5% | 74.4% | 91.9% | 58.1% |
| Meta Muse Spark Open | 80.3% | 76.7% | 90.6% | 72.6% | 82.6% | 85.7% |
| DeepSeek V4.1 Flash 1017 | 79.5% | 86.2% | 83.7% | 60.0% | 81.6% | 89.0% |
| Google Gemini 3.8 Flash | 79.1% | 79.7% | 86.7% | 72.3% | 80.4% | 77.6% |
| MoonshotAI Kimi K3 | 79.0% | 78.4% | 95.1% | 67.7% | 69.2% | 87.5% |
| Xiaomi MiMo-V2.6-Pro | 76.7% | 77.0% | 77.4% | 64.3% | 80.7% | 91.6% |
| Google Gemini 3.7 Flash | 76.5% | 83.3% | 89.7% | 63.6% | 89.4% | 51.7% |
| Qwen3.8 Flash | 74.6% | 68.2% | 92.6% | 71.9% | 54.5% | 90.3% |
| Z.ai GLM 5.3 | 73.2% | 89.8% | 74.1% | 32.0% | 93.2% | 81.6% |
| StepFun Step 5 Preview | 72.8% | 67.3% | 69.6% | 71.2% | 81.4% | 84.3% |
| NVIDIA Nemotron 3.5 Super VL | 70.9% | 64.4% | 77.5% | 69.5% | 65.6% | 84.9% |
| Z.ai GLM 5.3 Flash | 70.1% | 72.1% | 69.3% | 56.0% | 70.7% | 89.4% |
| Qwen3.8 Max (0902) | 69.9% | 73.0% | 90.1% | 57.8% | 71.0% | 55.0% |
| OpenAI GPT-6 Sol | 69.3% | 83.8% | 89.3% | 46.7% | 57.9% | 57.4% |
| DeepSeek V4 Flash Vision Exp | 69.2% | 80.6% | 83.0% | 36.2% | 67.0% | 81.7% |
| Meta Muse Spark 1.3 | 68.6% | 87.9% | 98.4% | 28.5% | 60.7% | 58.2% |
| MiniMax M3 | 68.0% | 73.7% | 63.6% | 62.7% | 76.9% | 60.1% |
| Meta Muse Spark 1.2 | 67.7% | 73.3% | 96.6% | 50.1% | 55.4% | 57.5% |
| Anthropic Claude Sonnet 4.6 | 63.6% | 71.8% | 77.1% | 60.2% | 42.8% | 53.0% |
| OpenAI GPT-5.5 | 61.6% | 66.7% | 74.2% | 53.6% | 55.1% | 52.5% |
| OpenAI GPT-5.6 Terra | 60.4% | 57.8% | 71.7% | 40.6% | 56.0% | 88.6% |
| Thinking Machines Inkling | 59.9% | 57.8% | 60.7% | 60.0% | 64.9% | 58.9% |
| Qwen3.8 27B | 59.4% | 63.5% | 76.5% | 37.9% | 59.4% | 62.8% |
| OpenAI GPT-5.6 Sol | 54.5% | 66.2% | 80.9% | 31.2% | 58.9% | 26.2% |
| OpenAI GPT-5.6 Luna | 51.0% | 65.6% | 25.0% | 29.3% | 83.6% | 55.3% |
| Google Gemma 4 31B | 44.4% | 49.3% | 40.8% | 48.5% | 47.7% | 27.2% |
| OpenAI GPT-6 Luna | 43.5% | 35.9% | 73.4% | 14.3% | 55.3% | 58.3% |
| Anthropic Claude Haiku 4.5 | 42.2% | 43.5% | 48.0% | 45.3% | 39.4% | 29.1% |
| Mistral Large 4 | 28.1% | 33.5% | 45.6% | 11.0% | 48.7% | 0.0% |
The parts as LiveBench publishes them, for the same run as the result.