LiveNerf
Repeated question-panel evaluations through Claude Code. Client version pinned; subscription access.
Data & methodology ↗Sources
Evaluations belong to their original authors. Each chart links to its data and methodology.
MarginLab · Claude Code
Daily SWE-Bench-Pro pass rates. Model periods follow MarginLab’s published transition dates; older models are shown separately.
Historical data ↗MarginLab · Codex
Daily SWE-Bench-Pro pass rates. Model periods follow MarginLab’s published transition dates; older models are shown separately.
Historical data ↗Additional trackers
Crowd Radar
Community coding evaluations. Unlink imports the published hourly score snapshots for available model series.
Claude Code view ↗ModelRegression
Multi-model benchmark history. Unlink imports the source’s published composite scores.
Compare models ↗DailyBench
Multiple API models · Opus 5.5 coverage not verified
A separate evaluation project. No current Opus 5.5 series verified; not included in our Opus charts.
Open tracker ↗AI Drift Detector
Frontier models · Opus 5.5 coverage not verified
External tracker, not a connected data feed. Its coverage and alert definitions differ from our sources.
Open tracker ↗Is It Cooked?
Multiple models · no Opus 5.5 series verified
The page showed a last run of August 10, 2026 when last reviewed. Treat this as a stale source until refreshed.
Open tracker ↗