Claude Opus 4

Claude Opus 4 (API name claude-opus-4-20250514) launched on 22 May 2025 as the top half of the Claude 4 generation, and Anthropic called it the world's best coding model. The benchmark behind that claim was 72.5% on SWE-bench Verified and 43.2% on Terminal-bench, but the number that mattered more was not a benchmark at all: Rakuten ran the model on an open-source refactor that continued independently for seven hours without performance falling off. Until then, agentic coding tools were judged on whether they could finish a single task; from here the question became how long a model could keep working before it lost the thread. The generation introduced extended thinking with tool use in beta — the model could search the web or call tools in the middle of its own reasoning rather than only before or after it — plus parallel tool calls and, where developers granted access to local files, the ability to write facts down and pick them up later. Claude Code left research preview and became generally available on the same day, with GitHub Actions, VS Code and JetBrains support. Pricing stayed at the established Opus rate of $15 per million input tokens and $75 per million output, inside a 200,000-token context window with up to 32,000 output tokens. Opus 4 was also the first Anthropic model released under ASL-3 deployment safeguards, a step up in the company's own risk framework rather than a marketing line. That matters to anyone reading the history of frontier-model safety practice: the threshold moved with this release, not with a policy document. Anthropic notified developers on 14 April 2026 and retired both Claude 4 models on 15 June 2026. The recommended replacement is claude-opus-4-8. Anthropic has committed to preserving the weights of retired models rather than deleting them.

▸Videos

No videos yet.

▸Reviews