← Claude Sonnet 5 · API · High effort (default)
Evidenceacceptedvendor_reported
Humanity's Last Exam
Claude Sonnet 5 System Card — Humanity's Last Exam (with tools)
ev_2026_07_sonnet5_hle
Result
57.4%
Confidence B
- Tested
- 2026-06-30
- Runs
- 1
Configuration
How it was measured
- Exact model
- claude-sonnet-5
- Reasoning
- high
- Agent harness
- none_or_vendor_hle_tools
- Tool access
- tools enabled (vendor config)
- Prompting
- vendor HLE protocol with tools enabled
SECONDARY / vendor tool-enabled HLE. Prefer independent AA HLE (ev_2026_07_sonnet5_aa_hle, ~39.6%) for closed-ended comparisons; this row is tool-assisted and not directly comparable.
Caveats
Known limitations
- Vendor-reported; independent HLE boards may differ by tool policy and grader.
- Grader model updates have restated prior Sonnet scores historically.
- Tool-enabled HLE is not pure closed-book reasoning.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.90
quality0.55
recency1.00
health0.90
comparability0.80
reliability0.65
Benchmark
Humanity's Last Exam
Frontier closed-ended academic benchmark across many subjects; often reported with and without tools.
Benchmark source →