PhaseoPhaseo
ModelsChatCompareAppsRankings

Help keep public data accurate

See something missing, incorrect, or out of date? Tell us about this page.

PhaseoPhaseo
Checking statusChecking statusVisit status page
Component-level status is unavailable.

Explore

  • Models
  • Chat
  • Providers
  • Apps
  • Rankings
  • Tools

Resources

  • Compare
  • Migration Guides
  • Methodology
  • Blog

Community

  • Discord
  • GitHub
  • LinkedIn
  • Reddit
  • X

Build

  • Documentation
  • API Reference
  • Quickstart
  • SDKs

Company

  • About
  • Trust Centre
  • Mission
  • Pricing
  • Works With
  • Acknowledgements
  • Support
  • Privacy
  • Terms

Explore

  • Models
  • Chat
  • Providers
  • Apps
  • Rankings
  • Tools

Build

  • Documentation
  • API Reference
  • Quickstart
  • SDKs

Resources

  • Compare
  • Migration Guides
  • Methodology
  • Blog

Company

  • About
  • Trust Centre
  • Mission
  • Pricing
  • Works With
  • Acknowledgements
  • Support
  • Privacy
  • Terms

Community

  • Discord
  • GitHub
  • LinkedIn
  • Reddit
  • X

© 2025 • Phaseo

Change language: English (UK)
  • English (UK)
  • 简体中文
  • हिन्दी
  • Español (España)
  • Français (France)
  • Deutsch (Deutschland)
  • Português (Brasil)
  • 日本語
  • العربية

Help:Issue·Support

Need help with Phaseo?Open an issueorcontact support

OSWorld

OSWorld

AI benchmark results and model performance

Type: percentage
View benchmark source

Recorded results

53

Average score

47.31%

Score range

5.92% - 85%

Leading model

85% - Claude Fable 5
Scores over time

Individual benchmark scores plotted by date.

Model results

OrganisationModelsReportedTop scoreInfoSelf reportedSource
AnthropicAnthropic
Anthropic
Claude Fable 509 Jun 202685%OSWorld-Verified; higher of Mythos 5 / Fable 5 in Anthropic's launch tableYesSource
AnthropicAnthropic
Anthropic
Claude Mythos 509 Jun 202685%OSWorld-Verified; higher of Mythos 5 / Fable 5 in Anthropic's launch tableYesSource
AnthropicAnthropic
Anthropic
Claude Opus 4.828 May 202683.40%OSWorld-Verified; updated Anthropic evaluation setupYesSource
AnthropicAnthropic
Anthropic
Claude Sonnet 530 Jun 202681.20%OSWorld-Verified; pass@1 averaged over five runs; 361 tasks, 100 steps; adaptive thinking at max effortYesSource
Meta
Meta
Muse Spark 1.109 Jul 202680.80%OSWorld-Verified; agentic computer useYesSource
AnthropicAnthropic
Anthropic
Claude Mythos Preview07 Apr 202679.60%VerifiedYesSource
OpenAIOpenAI
OpenAI
GPT 5.523 Apr 202678.70%VerifiedYesSource
AnthropicAnthropic
Anthropic
Claude Opus 4.716 Apr 202678%OSWorld-VerifiedYesSource
OpenAIOpenAI
OpenAI
GPT 5 Pro07 Aug 202575%inferred family alias from gpt-5.4 (score=0.4083; benches=19)YesSource
OpenAIOpenAI
OpenAI
GPT 5 Search API14 Oct 202575%inferred family alias from gpt-5.4 (score=0.3050; benches=19)YesSource
OpenAIOpenAI
OpenAI
GPT 5.405 Mar 202675%VerifiedYesSource
AnthropicAnthropic
Anthropic
Claude Opus 4.605 Feb 202672.70%OSWorld-VerifiedYesSource
AnthropicAnthropic
Anthropic
Claude Sonnet 4.617 Feb 202672.50%OSWorld-VerifiedYesSource
OpenAIOpenAI
OpenAI
GPT 5.4 Mini17 Mar 202672.10%VerifiedYesSource
MiniMax
MiniMax
MiniMax M301 Jun 202670%VerifiedYesSource
Qwen
Qwen
Qwen 3 VL 235B A22B Instruct-66.70%-Yes-
AnthropicAnthropic
Anthropic
Claude Opus 4.524 Nov 202566.26%Pass@1; Avg@5, 64k ThinkingYesSource
OpenAIOpenAI
OpenAI
GPT 5.3 Codex05 Feb 202664.70%OSWorld-Verified; xhigh reasoningYesSource
Qwen
Qwen
Qwen 3.6 Plus01 Apr 202662.50%-YesSource
z.AIz.AI
z.AI
GLM 5V Turbo01 Apr 202662.30%-YesSource
Qwen
Qwen
Qwen 3.5 122B A10B24 Feb 202658%-YesSource
Qwen
Qwen
Qwen 3.5 27B24 Feb 202656.20%-YesSource
Qwen
Qwen
Qwen 3.5 Flash23 Feb 202656.20%inferred family alias from qwen3.5-27b (score=0.4147; benches=81)YesSource
Qwen
Qwen
Qwen 3.5 35B A3B24 Feb 202654.50%-YesSource
AnthropicAnthropic
Anthropic
Claude Haiku 4.515 Oct 202550.70%inferred alias from claude-haiku-4-5-20251001YesSource
Report an issue
Sign Up