Revision History
Claude Sonnet 4.6 · 12 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Correction: Corrected GPQA Diamond (89.9%, not 74.1%, so no regression) and MMMLU (89.3% was published), removed a SWE-bench Pro score that belonged to GPT-5.3 Codex and tau2-bench per-domain scores that belonged to Sonnet 4.5, fixed the BrowseComp revision figures, the advisor-strategy savings (11.9% versus Sonnet alone, with Opus 4.6), the API default effort (high), prompt-caching minimums, the Snowflake partnership date (December 2025) and Salesforce model (Sonnet 4.5), reattributed the 94% insurance result to Pace, corrected two job titles, and other errors (plus 58 minor fixes)
Version 12 · Nov 2, 2026, 11:06 AM
- Improve internal links (2026-10-13)
Version 10 · Oct 13, 2026, 10:21 AM
- Correction: tau2-bench added a dual-control telecom domain; voice and knowledge retrieval are tau3-bench
Version 9 · Oct 1, 2026, 02:26 PM
- Correction: renamed 'Sierra Research' (a GitHub org slug) to Sierra, and repointed two Tau-Bench links from the tau_bench redirect to the canonical tau-bench slug.
Version 8 · Oct 1, 2026, 02:26 PM
- Correction: the Microsoft 365 E7 general availability post lists Copilot, E5, Entra Suite and Agent 365; Copilot Cowork was not part of the GA bundle and was in research preview when E7 was announced
Version 7 · Sep 30, 2026, 03:21 AM