← Claude Sonnet 5 · API · High effort (default)
Evidenceacceptedindependent_lab
LiveBench
LiveBench 2026-06-25 — Claude Sonnet 5 xHigh (Instruction Following category)
ev_2026_07_sonnet5_livebench_if
Result
Configuration
How it was measured
- Exact model
- claude-sonnet-5
- Reasoning
- max
- Agent harness
- none
- Tool access
- LiveBench default
- Prompting
- LiveBench standard
LiveBench Instruction Following category (stretch CHA). Category order on the public board is interpreted as Avg · Reasoning · Coding · Math · Data Analysis · Language · IF · [optional 8th]. Re-verify column headers at freeze.
Caveats
Known limitations
- Column mapping confidence medium until headers re-verified.
- xHigh effort.
- IFBench (harder precise-IF) is null for Sonnet 5 on AA as of scrape — do not invent IFBench.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.90
quality0.90
recency0.95
health0.90
comparability0.80
reliability0.75
Benchmark
LiveBench
Contamination-resistant live benchmark suite with category scores including reasoning, coding, mathematics, data analysis, language, and instruction following.
Benchmark source →