Claude Opus 4.5

Claude Opus 4.5 is Anthropic's late-2025 flagship and the model that made "80% on SWE-bench Verified" a reality: 80.9% on the benchmark, first place across frontier models at launch, plus leads on SWE-bench Multilingual in 7 of 8 languages.

The other headline is efficiency: a new effort parameter lets developers trade reasoning depth for cost — at medium effort it matches Sonnet 4.5's best SWE-bench score with 76% fewer output tokens, and at max effort it beats Sonnet 4.5 by 4.3 points using 48% fewer tokens (Anthropic-reported). It also leads on computer use (OSWorld) and agent benchmarks (τ2-bench).

Priced at $5/$25 per 1M input/output tokens — a 66% cut from Opus 4.1's $15/$75 — it became Anthropic's workhorse for Claude Code, Cursor, GitHub Copilot, and enterprise agents. It has since been superseded by Opus 4.6/4.7 at the same price, but remains the historical reference point for the 80% barrier.

Provider-reported (Anthropic, launch announcement):

• SWE-bench Verified: 80.9% — first model above 80%
• SWE-bench Multilingual: leads in 7 of 8 languages
• OSWorld (computer use): 66.3% — best-in-class at launch
• Medium effort: matches Sonnet 4.5's best SWE-bench with 76% fewer output tokens

Third-party coverage: TokenMix (2026-04-25) verified the figures against Anthropic's release notes and documents the 4.x family evolution; broad press coverage (CNBC, TechCrunch, Nov 2025) corroborates the launch, pricing cut, and coding-leadership claims. No independent re-run of Anthropic's exact benchmark numbers was found.