LiveBench 2.0 Data Analysis
Higher is better
- What does it measure?
- Tests whether a model answers short data-analysis questions on small tables correctly.
- Who evaluates it, and how?
- Questions with a single correct answer, such as finding runs of consecutive growth, mapping columns between two tables, and converting tables between CSV, JSON, Markdown and other formats, answered without tools (150 questions in the October 25, 2026 release). Answers are checked against ground truth.
- Reading the score
- Scores are shown as published by the source. Higher is better.
Top score85.2 / 100Gemini 4 Argon
Models tested22
Last updated2026.11.06
Model release date (newest on the right)
Best score so farHigher on the chart is better
Can it pull scattered facts from several long documents into one answer?Long Context Reasoning
Kimi K3🥇Kimi K3🥇Which AI gives the business answers people prefer most?Arena Business, Management & Finance
Gemini 4 Argon🥇Gemini 4 Argon🥇
Kimi K3🥇Kimi K3🥇Which AI gives the business answers people prefer most?Arena Business, Management & FinanceGemini 4 Argon🥇Gemini 4 Argon🥇- 🥇GoogleGemini 4 ArgonReasoning effort: High79.7%85.2 / 1001517
- 🥈OpenAIGPT-6 AstraReasoning effort: Max80.7%83.0 / 1001478
- 🥉OpenAIGPT-6.1 SolReasoning effort: Max84.0%82.7 / 1001483
- #4OpenAIGPT-6.1 TerraReasoning effort: Max82.7%82.3 / 100—
- #5AnthropicClaude Fable 5.2Reasoning effort: Extra High—81.5 / 100—
- #6AnthropicClaude Fable 5.1Reasoning effort: Max85.3%80.3 / 1001489
- #6AnthropicClaude Opus 5.5Reasoning effort: Max84.7%80.3 / 1001493
- #8MetaMuse Spark 1.3Reasoning effort: Max83.0%79.6 / 1001497
- #9DeepSeekDeepSeek V4.1 Flash 1017Reasoning effort: Max82.7%79.5 / 100—
- #10OpenAIGPT-5.6 TerraReasoning effort: Max83.0%79.3 / 1001466
- #11Moonshot AIKimi K3.1 ProReasoning effort: Max86.3%79.1 / 1001501
- #12Moonshot AIKimi K3Reasoning effort: Max88.7%78.7 / 1001488
- #13SpaceXSIGrok 4.7Reasoning effort: Extra High80.0%76.9 / 1001437
- #14Mistral AIMistral Large 4Reasoning effort: High81.3%76.5 / 1001449
- #15Z.aiGLM 5.3 FlashReasoning effort: Max80.0%76.4 / 1001471
- #16SpaceXSIGrok 4.6Reasoning effort: Extra High81.0%75.3 / 1001446
- #17OpenAIGPT-6 LunaReasoning effort: Max83.3%73.4 / 1001440
- #18XiaomiMiMo-V2.6-ProReasoning effort: High86.3%72.3 / 1001460
- #19AnthropicClaude Sonnet 5.5Reasoning effort: Max82.7%59.5 / 1001472
- #20GoogleGemini 3.8 FlashReasoning effort: High84.0%54.0 / 1001486
- #21GoogleGemma 4 31BReasoning effort: Max69.7%53.3 / 1001449
- #22AnthropicClaude Haiku 5.5Reasoning effort: Max82.7%53.1 / 1001431
Half of models ≤ 78.9 / 100Last updated 2026-11-11
The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.
Where the scores come from
- Run by the test’s team: scores the team that operates the test ran and published itself.