Claude Opus 4.5 Arrives: State-of-the-Art Coding, a New Effort Parameter, and a Steep Price Cut
Anthropic has released Claude Opus 4.5, the newest addition to its Opus line and, by the company's own benchmarks, its strongest model yet for coding, agentic workflows, and computer use. Opus 4.5 posts state-of-the-art results on SWE-bench Verified among frontier models, a 10.6% improvement over Sonnet 4.5 on the Aider Polyglot coding benchmark, and 29% more points than Sonnet 4.5 on Vending-Bench, a long-horizon agentic test. In an internal engineering take-home exam given a strict two-hour limit, Opus 4.5 outperformed every human candidate who took the same test.
A new dial for cost vs. capability
Opus 4.5 introduces an effort parameter in the API, letting developers trade thinking depth for token spend on a sliding scale rather than an all-or-nothing switch:
- Medium effort matches Sonnet 4.5's best recorded score while using 76% fewer output tokens.
- Highest effort exceeds Sonnet 4.5's best score by 4.3 percentage points while still using 48% fewer tokens than Sonnet 4.5 needed.
Effort has historically been bundled into which model you pick. Splitting it into its own parameter means a team can stay on Opus 4.5 for consistency and just turn the dial down for routine agentic steps — rather than routing cheap tasks to a smaller model and hoping quality doesn't suffer.
Cheaper, and everywhere at once
Pricing drops to $5 per million input tokens and $25 per million output tokens, down from $15/$75 for Opus 4.1 — a two-thirds cut. Opus 4.5 (API model ID claude-opus-4-5-20251101) is available today in the Claude apps, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry simultaneously — the broadest same-day platform rollout of any Claude model to date.