Anthropic deprecates Claude Opus 4.1, retiring it from the API on August 5, 2026

Anthropic announced on June 5, 2026 that Claude Opus 4.1 (claude-opus-4-1-20250805) is deprecated, with retirement from the Claude API scheduled for August 5, 2026. Teams still using Opus 4.1 have two months to migrate. Anthropic recommends migrating to Claude Opus 4.8, which launched on May 28, 2026.

What's new

• Model affected: claude-opus-4-1-20250805
• Deprecation announced: June 5, 2026
• Retirement date: August 5, 2026
• Recommended migration target: Claude Opus 4.8 (claude-opus-4-8)
• Migration guide: Available in Anthropic's documentation at the model deprecations page

Context

Claude Opus 4.1 was released on August 5, 2025. Its retirement date of August 5, 2026 marks exactly one year from initial launch — consistent with Anthropic's general pattern of granting GA models roughly a year before deprecation begins.

With Claude Opus 4.8 now in general availability as of May 28, 2026, Anthropic is consolidating the active Opus tier. Opus 4.8 supersedes Opus 4.1 on every major dimension: 1M token context window, lower prompt caching threshold, adaptive thinking, mid-conversation system messages, and the same $5/$25 per million token pricing. There is no cost or feature penalty for migrating.

Anthropic's deprecation cadence has been consistent: preview models retire within 3–6 months of the stable version going GA; GA models receive approximately one year of active support before deprecation. Opus 4.1 followed the GA path.

Why it matters

For most teams, the migration from Opus 4.1 to Opus 4.8 is low-friction: same pricing, same API interface, same tool support. The main breaking change is that the sampling parameters temperature, top_p, and top_k return a 400 error on non-default values in Opus 4.8 — behavior introduced in Opus 4.7 that carries over. Teams using non-default sampling values will need to remove those parameters.

The other migration consideration is the effort parameter, which defaults to high on Opus 4.8. Teams that relied on the lower default effort of Opus 4.1 to control latency may need to explicitly set effort: "medium" or effort: "low" to match prior behavior and costs.

August 5 is the hard deadline. After that date, calls to claude-opus-4-1-20250805 will fail. Teams running Opus 4.1 in production should prioritize this migration over the next two months.