Home Model series Claude Sonnet Claude Sonnet · CharXiv Reasoning · Validation split · Release history Claude Sonnet · CharXiv Reasoning · Validation split · Release history View scores, configurations, and disclosures for CharXiv Reasoning · Validation split across 2 Claude Sonnet releases, ordered by release date.
Differences are calculated only for unique, active, exact values between direct predecessors when the unit, metric, and explicitly recorded configuration and conditions match.
2026-02-17 · 3 score records for this release, including 3 exact values (may include historical records)
Claude Sonnet 4.6 · CharXiv Reasoning scores and configurations Metric and score Configuration and evaluation conditions Sources Accuracy 72.4% Reported 72.4% · Unit % Published Unconfirmed
Adaptive max, without tools Settings Adaptive thinking enabled; max effort; no tools. Unknowns Default sampling settings apply unless otherwise stated. Tools Tools forbidden Reasoning budget max Framework not recorded Protocol CharXiv Reasoning, 1,000 validation questions; five-run average. Tools: none. Conditions Without tools; adaptive thinking, max effort. View this score in rankings Accuracy 77.4% Reported 77.4% · Unit % Published Unconfirmed
Conflicting results are pending review and are excluded from exact rankings.
Adaptive max, image-cropping tool Settings Adaptive thinking enabled; max effort; simple image-cropping tool available. Unknowns Default sampling settings apply unless otherwise stated. Tools Tools allowed Reasoning budget max Framework not recorded Protocol CharXiv Reasoning, 1,000 validation questions; five-run average. Tools: image cropping. Conditions Simple image-cropping tool; adaptive thinking, max effort. 同一系统卡图中为 79.4%,正文为 77.4%;未发现更正依据,两值均不参与精确排名。 View this score in rankings Accuracy 79.4% Reported 79.4% · Unit % Published Unconfirmed
Conflicting results are pending review and are excluded from exact rankings.
Adaptive max, image-cropping tool Settings Adaptive thinking enabled; max effort; simple image-cropping tool available. Unknowns Default sampling settings apply unless otherwise stated. Tools Tools allowed Reasoning budget max Framework not recorded Protocol CharXiv Reasoning, 1,000 validation questions; five-run average. Tools: image cropping. Conditions Figure 2.17.3.A prints 79.4% for Sonnet 4.6 with an image-cropping tool, conflicting with the 77.4% prose result in §2.17.3, PDF p.36, already recorded in the core candidate. Both source values are retained without resolving the discrepancy. View this score in rankings
§2.17.3, PDF p.36 Attachment · Claude Sonnet 4.6 System Card Figure 2.17.3.A, PDF p.37, Sonnet 4.6 bars Attachment · Claude Sonnet 4.6 System Card 2026-06-30 · 2 score records for this release, including 2 exact values (may include historical records)
Records for the direct predecessor do not provide matching conditions for a valid difference.
Claude Sonnet 5 · CharXiv Reasoning scores and configurations Metric and score Configuration and evaluation conditions Sources Accuracy 77% Reported 77.0% · Unit % Published Unconfirmed
Adaptive thinking, max effort, no tools Settings Adaptive thinking at max effort; no tools. Unknowns none recorded Tools Tools forbidden Reasoning budget max Framework not recorded Protocol CharXiv Reasoning, 1,000 validation questions; five-run average. Tools: none. Conditions No tools; adaptive thinking and max effort; Sonnet 4.6 grader. View this score in rankings Accuracy 88.3% Reported 88.3% · Unit % Published Unconfirmed
Adaptive thinking, max effort, tools enabled Settings Adaptive thinking at max effort; tools enabled as specified per benchmark. Unknowns none recorded Tools Tools allowed Reasoning budget max Framework not recorded Protocol CharXiv Reasoning, 1,000 validation questions; five-run average. Tools: Python and image cropping. Conditions Python tools and image-cropping available; adaptive thinking and max effort; Sonnet 4.6 grader. View this score in rankings
§8.10.5, PDF p.129 Attachment · Claude Sonnet 5 System Card