AI rankings for overall performance

LiveBench 2.0

Often citedHigher is better
What does it measure?
The overall score of LiveBench 2.0, combining seven suites into one: long agentic work, coding, browser use, design, research, data analysis and reasoning.
Who evaluates it, and how?
A weighted mean of the suites the model has a score for. The October 25, 2026 release weights Long Agentic ×2, Coding and Design ×1.5 and the rest ×1; earlier releases weight every suite equally.
Reading the score
Scores are shown as published by the source. Higher is better. LiveBench 2.0 has different suites and questions from v1, so scores are not comparable with v1 (v1 scores remain in the fields marked 'v1').
Other versions:LiveBench 2.0 BrowserLiveBench 2.0 CodingLiveBench 2.0 Data AnalysisLiveBench 2.0 DesignLiveBench 2.0 Long AgenticLiveBench 2.0 ReasoningLiveBench 2.0 Research
Top score84.9 / 100Claude Fable 5.2
Models tested22
Last updated2026.11.06
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 🥇AnthropicClaude Fable 5.2Reasoning effort: Extra High64.6—84.9 / 100
  2. 🥈OpenAIGPT-6 AstraReasoning effort: Max56.9147484.5 / 100
  3. 🥉AnthropicClaude Fable 5.1Reasoning effort: Max50.0150384.0 / 100
  4. #4AnthropicClaude Opus 5.5Reasoning effort: Max59.1150683.4 / 100
  5. #5OpenAIGPT-6.1 SolReasoning effort: Max54.6148281.4 / 100
  6. #6OpenAIGPT-6.1 TerraReasoning effort: Max51.3—78.8 / 100
  7. #7GoogleGemini 4 ArgonReasoning effort: High50.4152277.7 / 100
  8. #8AnthropicClaude Sonnet 5.5Reasoning effort: Max54.5147377.0 / 100
  9. #9DeepSeekDeepSeek V4.1 Flash 1017Reasoning effort: Max32.6—70.3 / 100
  10. #10AnthropicClaude Haiku 5.5Reasoning effort: Max36.5142269.0 / 100
  11. #11Moonshot AIKimi K3.1 ProReasoning effort: Max46.5148368.7 / 100
  12. #12MetaMuse Spark 1.3Reasoning effort: Max36.6149668.4 / 100
  13. #13SpaceXSIGrok 4.7Reasoning effort: Extra High35.5144267.9 / 100
  14. #14OpenAIGPT-5.6 TerraReasoning effort: Max32.3146867.2 / 100
  15. #15XiaomiMiMo-V2.6-ProReasoning effort: High29.9148066.1 / 100
  16. #16Moonshot AIKimi K3Reasoning effort: Max17.4148765.0 / 100
  17. #17SpaceXSIGrok 4.6Reasoning effort: Extra High31.0145763.5 / 100
  18. #18Z.aiGLM 5.3 FlashReasoning effort: Max27.8147460.7 / 100
  19. #19OpenAIGPT-6 LunaReasoning effort: Max27.9144160.4 / 100
  20. #20GoogleGemini 3.8 FlashReasoning effort: High30.8149856.0 / 100
  21. #21Mistral AIMistral Large 4Reasoning effort: High25.3142340.9 / 100
  22. #22GoogleGemma 4 31BReasoning effort: Max4.4145232.4 / 100
Half of models ≤ 68.5 / 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.