Frontier Log

Claude Sonnet 5 vs Claude Sonnet 5.5

Claude Sonnet 5Claude Sonnet 5.5
Released
LineupFable › Opus › Sonnet › HaikuFable › Opus › Sonnet › Haiku
Input / 1M tokens$2$2
Cached input / 1M tokens$0.20$0.10
Output / 1M tokens$10$10
Context window1M1M
Output limit128K128K
Benchmarks
Humanity's Last Exam No tools, full set (2,500 questions)43.2%56.9%
Humanity's Last Exam With tools, full set (2,500 questions)57.4%64.5%
SWE-bench Pro Set not specified63.2%81.3%1
SWE-bench Pro V2 Full set (642 tasks)93.15%2Claude Code95.79%3mini-SWE-agent
SWE-bench Pro V2 HARD-51 (51 tasks)88.20%4Claude Code70.59%5mini-SWE-agent
Terminal-Bench 4.112.1%663.9%7
AutomationBench 1.0.6 Private set10.7%844.7%9
OSWorld 2.1 Full set (108 tasks), partial score57.0%1080.1%11
OSWorld 2.1 Full set (108 tasks), strict pass rate25.6%1243.5%13
Chartography No tools16.0%1461.6%
Chartography With tools71.8%1590.2%16
ExploitGym Mitigations off, successful exploits51 of 86917231 of 86918

Download image

Claude Sonnet 5Claude Sonnet 5.5
Other benchmarks
BrowseComp84.7%19No data
Terminal-Bench 4.0No data70.6%20
Terminal-Bench Science 0.1No data59.9%21
AutomationBench, release not specified Private set13.5%22No data
OSWorld 2.1 Offline set (82 tasks), partial scoreNo data83.9%23
OSWorld 2.1 Offline set (82 tasks), strict pass rateNo data48.8%24
FrontierCode 1.1 Main setNo data52.1%25
FrontierCode 1.1 Extended setNo data64.4%26
FrontierCode 2.0 Score, all tasks (132)No data42.2%27
FrontierCode 2.0 Pass rate, all tasks (132)No data46.4%28
FrontierSWE v2No data61.9%29
DeepSWE v1.1No data71.0%30
Terminal-Bench 2.180.4%No data
OSWorld-Verified81.2%31No data

Notes

  1. SWE-bench Pro, Set not specified, Claude Sonnet 5.5: max effort · adaptive thinking · average of 5 trials
  2. SWE-bench Pro V2, Full set (642 tasks), Claude Sonnet 5: Scale AI’s leaderboard, 18 Nov 2026 · Claude Code · xhigh effort
  3. SWE-bench Pro V2, Full set (642 tasks), Claude Sonnet 5.5: Scale AI’s leaderboard, 18 Nov 2026 · mini-swe-agent · xhigh effort
  4. SWE-bench Pro V2, HARD-51 (51 tasks), Claude Sonnet 5: Scale AI’s leaderboard, 18 Nov 2026 · Claude Code · xhigh effort
  5. SWE-bench Pro V2, HARD-51 (51 tasks), Claude Sonnet 5.5: Scale AI’s leaderboard, 18 Nov 2026 · mini-swe-agent · xhigh effort
  6. Terminal-Bench 4.1, Claude Sonnet 5: Stanford University and Laude Institute’s leaderboard, 18 Nov 2026 · Claude Code · max effort
  7. Terminal-Bench 4.1, Claude Sonnet 5.5: Stanford University and Laude Institute’s leaderboard, 18 Nov 2026 · Claude Code · max effort
  8. AutomationBench 1.0.6, Private set, Claude Sonnet 5: Run by Zapier
  9. AutomationBench 1.0.6, Private set, Claude Sonnet 5.5: Run by Zapier · max effort · default fallbacks on
  10. OSWorld 2.1, Full set (108 tasks), partial score, Claude Sonnet 5: max effort · mean of 5 runs · September 10, 2026 task files
  11. OSWorld 2.1, Full set (108 tasks), partial score, Claude Sonnet 5.5: max effort · mean of 5 runs · September 10, 2026 task files
  12. OSWorld 2.1, Full set (108 tasks), strict pass rate, Claude Sonnet 5: max effort · mean of 5 runs · September 10, 2026 task files
  13. OSWorld 2.1, Full set (108 tasks), strict pass rate, Claude Sonnet 5.5: max effort · mean of 5 runs · September 10, 2026 task files
  14. Chartography, No tools, Claude Sonnet 5: max effort · adaptive thinking · average of 5 runs
  15. Chartography, With tools, Claude Sonnet 5: max effort · adaptive thinking · average of 5 runs · container and image-cropping tool
  16. Chartography, With tools, Claude Sonnet 5.5: max effort · adaptive thinking · average of 5 runs · container and image-cropping tool · Gemini 3.5 Flash grader
  17. ExploitGym, Mitigations off, successful exploits, Claude Sonnet 5: 6h wall-clock budget · agent judge confirms the intended vulnerability
  18. ExploitGym, Mitigations off, successful exploits, Claude Sonnet 5.5: 6h wall-clock budget · agent judge confirms the intended vulnerability
  19. BrowseComp, Claude Sonnet 5: max effort · adaptive thinking · 10M-token limit, compaction at 200K tokens
  20. Terminal-Bench 4.0, Claude Sonnet 5.5: adaptive thinking
  21. Terminal-Bench Science 0.1, Claude Sonnet 5.5: Claude Code in --bare mode · max effort · standard error ±4.7 points
  22. AutomationBench, release not specified, Private set, Claude Sonnet 5: Run by Zapier · max effort
  23. OSWorld 2.1, Offline set (82 tasks), partial score, Claude Sonnet 5.5: max effort · mean of 5 attempts
  24. OSWorld 2.1, Offline set (82 tasks), strict pass rate, Claude Sonnet 5.5: max effort · mean of 5 attempts
  25. FrontierCode 1.1, Main set, Claude Sonnet 5.5: Run by Cognition · Claude Code harness · xhigh effort
  26. FrontierCode 1.1, Extended set, Claude Sonnet 5.5: Run by Cognition · Claude Code harness · xhigh effort
  27. FrontierCode 2.0, Score, all tasks (132), Claude Sonnet 5.5: Cognition’s leaderboard, 18 Nov 2026 · Claude Code · max effort
  28. FrontierCode 2.0, Pass rate, all tasks (132), Claude Sonnet 5.5: Cognition’s leaderboard, 18 Nov 2026 · Claude Code · max effort
  29. FrontierSWE v2, Claude Sonnet 5.5: Run by Proximal · Proximal's agent harness · max reasoning effort · mean of 5 trials per task · 20-hour budget
  30. DeepSWE v1.1, Claude Sonnet 5.5: average of 5 trials
  31. OSWorld-Verified, Claude Sonnet 5: revised run method with zoom tool fix and 128K max tokens per turn

Related

All comparisons