Humanity's Last Exam No tools, full set · adaptive thinking, max effort; average of 5 trials 73.2% Humanity's Last Exam With tools, full set · highest available effort; safeguards on 74.2% Corroborated The same number is printed in 1 other lab document: 1 SWE-bench Pro Set not stated · adaptive thinking, max effort; average of 5 trials 93.0% Terminal-Bench 4.0 Claude Code in --bare mode; max effort; safeguards on; no internet access; 10 trials per task 67.6% Terminal-Bench Science 0.1 highest available effort; safeguards on 74.6% Corroborated The same number is printed in 1 other lab document: 1 AutomationBench 1.0.6 Private set · Run by Zapier · highest available effort; safeguards on; set and release 1.0.6 named in the system card, which prints the same figure 51.4% Corroborated The same number is printed in 1 other lab document: 1 Run by Zapier · low effort; default fallbacks on · in Claude Fable 5.2 and Claude Mythos 5.2 System Card 39.7% OSWorld 2.1 Offline set, partial score · highest available effort; safeguards on 89.3% Corroborated The same number is printed in 1 other lab document: 1 OSWorld 2.1 Offline set, strict pass rate · 82-task offline set, no internet access; max effort; mean of 5 attempts 54.6% DeepSWE v1.1 average of 5 trials 74.3% ARC-AGI-3 Set not stated · Run by ARC Prize Foundation · Standard harness; high effort 79.26% Run by ARC Prize Foundation · Provider Adapter harness; high effort 97.72% Run by ARC Prize Foundation · Standard harness; max effort 65.66% Run by ARC Prize Foundation · Provider Adapter harness; max effort 94.90% Chartography No tools · highest available effort; safeguards on 73.4% Corroborated The same number is printed in 1 other lab document: 1 medium effort 66.4% separate run for the redrawn-chart check; max effort; average of 5 runs · in Claude Fable 5.2 and Claude Mythos 5.2 System Card 74.3% Chartography With tools · adaptive thinking, max effort; container and image-cropping tool; average of 5 runs; Gemini 3.5 Flash grader 91.6%