All Benchmarks/Agents

LiveBench 2.0 Browser

Higher is better
What does it measure?
Tests whether a model can complete multi-step jobs in working web apps through a real browser.
Who evaluates it, and how?
The model works an online shop, project boards, a scheduler, an expense ledger, a helpdesk and a wiki, seeing them by screenshot (25 tasks in the October 25, 2026 release). The answer and the app's resulting data are checked against the expected outcome, and the score is the share of tasks passed.
Reading the score
Scores are shown as published by the source. Higher is better. A model that cannot take screenshots cannot do these tasks and scores 0.
Other versions:LiveBench 2.0LiveBench 2.0 CodingLiveBench 2.0 Data AnalysisLiveBench 2.0 DesignLiveBench 2.0 Long AgenticLiveBench 2.0 ReasoningLiveBench 2.0 Research
Top score93.3 / 100GPT-6.1 Sol
Models tested22
Last updated2026.11.06
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    OpenAI
    GPT-6.1 Sol
    2026.09.29 · max
    93.3 / 100
  2. 2
    OpenAI
    GPT-6.1 Terra
    2026.10.26 · max
    93.1 / 100
  3. 3
    OpenAI
    GPT-6 Astra
    2026.09.04 · max
    91.3 / 100
  4. 4
    Anthropic
    Claude Fable 5.1
    2026.09.01 · max
    89.8 / 100
  5. 5
    Anthropic
    Claude Sonnet 5.5
    2026.09.28 · max
    89.3 / 100
  6. 6
    Anthropic
    Claude Opus 5.5
    2026.09.22 · max
    88.4 / 100
  7. 7
    Anthropic
    Claude Fable 5.2
    2026.11.09 · xhigh
    86.9 / 100
  8. 8
    Google
    Gemini 4 Argon
    2026.09.30 · high
    82.8 / 100
  9. 9
    Xiaomi
    MiMo-V2.6-Pro
    2026.09.21 · high
    77.3 / 100
  10. 10
    Moonshot AI
    Kimi K3.1 Pro
    2026.10.23 · max
    75.9 / 100

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.