Best LLM for data analysis
- Gemini 4 Argon leads
- 85.2%
- models ranked
- 43
- lead over #2
- +2.2 pts
LiveBench data analysis · read October 22
As of October 22, 2026, Google Gemini 4 Argon leads 43 models on LiveBench data analysis with 85.2%, ahead of OpenAI GPT-6 Astra (83.0%) and OpenAI GPT-6.1 Sol (82.7%).
The full best LLM for general use ranking combines this benchmark with others.
| Rank | Model | LiveBench data analysis | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Gemini 4 Argon Google | 85.2% | 81.0 | $10.00 | |
| 2 | GPT-6 Astra OpenAI | 83.0% | 85.1 | $50.00 | |
| 3 | GPT-6.1 Sol OpenAI | 82.7% | 82.4 | $10.00 | |
| 4 | GPT-5.5 OpenAI | 81.6% | 71.3 | $30.00 | |
| 5 | GPT-6 Sol OpenAI | 81.2% | 76.0 | $10.00 | |
| 6 | Claude Fable 5 Anthropic | 80.5% | 79.3 | $50.00 | |
| 7 | Claude Fable 5.1 Anthropic | 80.3% | 84.2 | $50.00 | |
| 8 | Claude Opus 5.5 Anthropic | 80.3% | 88.8 | $20.00 | |
| 9 | GPT-5.6 Sol OpenAI | 79.8% | 73.5 | $20.00 | |
| 10 | Muse Spark 1.3 Meta | 79.6% | 57.4 | $4.25 | |
| 11 | DeepSeek V4 Flash Vision Exp DeepSeek | 79.5% | Not enough data | $1.32 | |
| 12 | DeepSeek V4.1 Flash 1017 DeepSeek | 79.5% | 70.3 | $0.60 | |
| 13 | DeepSeek V4.1 Flash DeepSeek | 79.3% | 62.5 | $0.84 | |
| 14 | GPT-5.6 Terra OpenAI | 79.3% | 63.9 | $12.00 | |
| 15 | Kimi K3 Moonshot AI | 78.7% | 62.3 | $15.00 | |
| 16 | Qwen3.8 Max (0902) Qwen | 78.4% | 62.4 | $6.00 | |
| 17 | GPT-5.6 Luna OpenAI | 78.0% | 58.0 | $1.20 | |
| 18 | Grok 4.7 xAI | 76.9% | 66.7 | $6.00 | |
| 19 | Qwen3.8 27B Qwen | 76.6% | 74.2 | $2.55 | |
| 20 | Muse Spark 1.2 Meta | 76.5% | 55.7 | $4.25 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their LiveBench data analysis result, as the board publishes it, best first. Short data-analysis questions on small tables (runs of consecutive growth, joining one table's columns onto another's, converting a table between formats), each with a single correct answer graded against ground truth. 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.
- OpenRouter, fetched
- LiveBench, fetched , board updated October 14, 2026