Claude Opus 4.1 retires on 5 August 2026: migrate to Opus 4.8 now

If your production code calls claude-opus-4-1-20250805, you have a hard deadline to work with. Anthropic announced the deprecation of that model on 5 June 2026, and it retires permanently on 5 August 2026. After that date, any request using that model ID returns an error. No graceful fallback, no warning response, just a failure.

The recommended migration target is claude-opus-4-8.

What happens on 5 August 2026

Anthropic’s model deprecations page is unambiguous: “requests to models past the retirement date will fail.” There is no grace period baked into the API. The moment the clock ticks over, claude-opus-4-1-20250805 stops responding.

Anthropic follows a 60-day notice window for publicly released models, which is exactly what happened here. The deprecation notice went out on 5 June 2026, giving teams two months to act. If you received that email and filed it away, that window is now closing fast.

Why this keeps catching teams out

Every Claude model ID is a pinned snapshot. The date in claude-opus-4-1-20250805 refers to the model’s release date, 5 August 2025, not a rolling alias. There is no evergreen pointer here. If you hardcoded that string anywhere, it breaks on retirement day.

The tricky part is that “anywhere” covers more ground than most people realise. The obvious place is your main application code. The less obvious places are:

Serverless functions that were set up once and rarely touched
CI/CD pipelines running evaluation or regression jobs
Fallback tiers in a routing layer that only activate under load
Archived microservices that are technically still running

If you deploy across both the direct Anthropic API and a cloud partner such as Amazon Bedrock or Google Cloud, check both surfaces separately. The deprecation calendars run on different clocks, and migrating on one track does not guarantee the other is clean.

What you are migrating to

Claude Opus 4.8 launched on 28 May 2026 and is Anthropic’s recommended replacement. The model ID is claude-opus-4-8.

The capability jump is meaningful. Opus 4.8 ships with a 1 million token context window by default, 128k maximum output tokens, and adaptive thinking built in. Anthropic reports it is around four times less likely than its predecessor to let code flaws pass unremarked, scores 84% on the Online-Mind2Web benchmark, and is the first model to break 10% on the Legal Agent Benchmark all-pass standard. For teams running agentic workloads, Workflows in Claude Code adds planning and parallel subagent support.

Pricing sits at $5 per million input tokens and $25 per million output tokens. A Fast mode is available at $10 input and $50 output, running roughly 2.5 times faster.

The model is available through the Claude Platform, Amazon Bedrock, Google Cloud, and Microsoft Foundry.

One parameter change to be aware of

If you are migrating from Opus 4.1, note that from the Opus 4.6 generation onwards, the temperature, top_p, and top_k parameters return an HTTP 400 error on non-default values. Anthropic expects behaviour steering to happen through prompting rather than sampling parameters. Check your API calls for those fields and remove them before you switch model IDs, otherwise you will swap one error for another.

How to check your exposure now

The fastest audit is a codebase-wide search for the string claude-opus-4-1. That catches the full model ID and any partial references in config files, environment variables, or infrastructure-as-code templates. From there:

Replace claude-opus-4-1-20250805 with claude-opus-4-8 in every location.
Remove any temperature, top_p, or top_k overrides from those calls.
Run your existing test suite against the new model ID in a non-production environment.
Check your cloud partner integrations separately if you use Bedrock or Vertex.

Anthropic’s models overview page lists the current active models and flags deprecated ones, so it is a useful reference to bookmark for future retirement cycles.

The broader pattern

Claude Sonnet 4 and Opus 4 both retired on 15 June 2026. Opus 4.1 follows on 5 August 2026. Anthropic is now running a deprecation cadence where models are retired roughly 60 to 90 days after their successor ships. That pace is not going to slow down. Building a process to track and action these notices, rather than treating each one as a one-off, will save time in the long run. The LLM Latency deprecation calendar is one useful third-party resource for tracking upcoming retirement dates across providers.

The migration itself is straightforward. The risk is not technical complexity; it is simply forgetting a hardcoded string somewhere in the stack until the deadline makes it impossible to overlook.