← Claude Opus 4.8 · API · High effort (default)
Evidenceacceptedindependent_lab
IFBench
Artificial Analysis — IFBench (Opus 4.8 max)
ev_2026_07_opus48_ifbench
Result
62.24%
Confidence A
- Tested
- 2026-06-01
- Runs
- 1
Configuration
How it was measured
- Exact model
- claude-opus-4-8
- Reasoning
- max
- Agent harness
- none
- Tool access
- none
- Prompting
- AA IFBench protocol
CHA support — AA IFBench (Opus has a published score; Sonnet was null).
Caveats
Known limitations
- Precise constraint IF; distinct from Arena preference.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.95
quality1.00
recency0.95
health0.90
comparability0.90
reliability0.85
Benchmark
IFBench
Harder verifiable instruction-following constraints designed to test generalization beyond saturated IFEval-style benchmarks.
Benchmark source →