← Claude Sonnet 5 · API · High effort (default)
Evidenceacceptedindependent_lab
LiveBench
LiveBench 2026-06-25 — Claude Sonnet 5 xHigh Effort (Reasoning category)
ev_2026_07_sonnet5_livebench
Result
Configuration
How it was measured
- Exact model
- claude-sonnet-5
- Reasoning
- max
- Agent harness
- none
- Tool access
- LiveBench default
- Prompting
- LiveBench standard
Independent LiveBench suite — use Reasoning category as INT support signal; Global Average is multi-skill and should not solely drive INT.
Caveats
Known limitations
- rawScore is the Reasoning category, not Global Average (74.8).
- xHigh effort is above typical default chat.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.85
quality0.95
recency0.95
health0.90
comparability0.85
reliability0.85
Benchmark
LiveBench
Contamination-resistant live benchmark suite with category scores including reasoning, coding, mathematics, data analysis, language, and instruction following.
Benchmark source →