Claude Sonnet 4.5

published

Anthropic's previous-gen balanced model

Strong coding model with hybrid reasoning. 77.2% on SWE-bench Verified standard, 82% with parallel compute.

Provider anthropic Type llm Access api Params undisclosed Context 200k License proprietary

Benchmarks (3)

Benchmark	Score	Reported By	Source
SWE-bench Verified	77.2%	vendor	source ↗
SWE-bench (parallel)	82.0%	vendor	source ↗
MMLU	89.1%	vendor	source ↗

First Claude model with hybrid extended thinking. Set SWE-bench records at release.

Superseded by Sonnet 4.6 and Opus 4.6 on most benchmarks.

Provider anthropic

Released 2025-09-01

Training cutoff 2025-04

Created March 22, 2026

Last reconciled Never