Decision first: As of November 12, 2026, Claude Sonnet 4.5 is deprecated but remains functional on Anthropic-operated platforms; Anthropic schedules its retirement for November 30, 2026 and recommends Claude Sonnet 5.5 as a replacement. That makes Sonnet 4.5 a poor choice for a new, long-lived Anthropic API deployment. Existing users should plan and test a migration before retirement. (Anthropic’s Sonnet 4.5 overview and model deprecations page)
These dates apply to Anthropic-operated platforms, including the Claude API. Partner-operated services set their own lifecycle schedules; for example, AWS lists a separate Bedrock end-of-life date. Check the schedule for the provider you use rather than assuming Anthropic’s date applies everywhere. (Anthropic’s model deprecations page; AWS Bedrock model card)
Model identity and production specifications
| Item | Anthropic Claude API |
|---|---|
| Dated model ID | claude-sonnet-4-5-20250929 |
| API alias | claude-sonnet-4-5 |
| Lifecycle | Deprecated; retires November 30, 2026 |
| Context window | 200K tokens |
| Maximum output | 64K tokens |
| Input and output | Text and images in; text out |
| Reliable knowledge cutoff | January 2025 |
| Standard price | $3 per million input tokens; $15 per million output tokens |
These specifications are for Anthropic’s Claude API; other providers may use different model IDs, prices, limits, or retirement schedules. (Anthropic’s Sonnet 4.5 overview)
Anthropic lists prompt-cache writes at $3.75 per million tokens for a five-minute cache and $6 per million tokens for a one-hour cache, with cache reads at $0.30 per million tokens. The Message Batches API offers a 50% discount on input and output for this model. Estimate costs using the features and channel your application will actually use. (Anthropic’s Sonnet 4.5 overview and pricing page)
Cost example: Across multiple requests, 1 million total input tokens plus 200,000 total output tokens costs $6 at standard rates: $3 for input and $3 for output. This base-price estimate excludes caching, batch processing, and other applicable terms.
What workloads is Sonnet 4.5 suited to?
Anthropic positioned Sonnet 4.5 for coding, complex agent tasks, and computer use. Its launch announcement reported a 61.4% OSWorld score and described the model maintaining focus for more than 30 hours on complex, multi-step tasks. These are Anthropic-reported results and observations—not a guarantee of how an agent will perform with your tools, prompts, or safeguards. Treat them as reasons to run a workload-specific trial, not as a substitute for one. (Anthropic’s launch announcement)
Anthropic’s Sonnet 4.5 System Card covers capability and safety evaluations, including agentic safety, computer use, and tool use. Consult it if your decision depends on evaluation scope or safety considerations; a short feature summary is not a substitute for the detailed document.
Practical fit: Before retirement, Sonnet 4.5 may still be relevant when maintaining an existing integration or comparing outputs during migration. For a new Anthropic API service expected to run beyond November 30, choose an active model instead. Anthropic recommends Sonnet 5.5 as the replacement. Its published specifications are a 1M-token context window, 128K maximum output, and base API rates of $2 per million input tokens and $10 per million output tokens. Those figures are a specification and price comparison—not evidence that Sonnet 5.5 will produce equivalent results for your workload. (Anthropic’s Sonnet 5.5 overview)
Migration checklist for existing API users
- Find every call site. Check application configuration, environment variables, scheduled jobs, and services that store the model ID. Anthropic’s Console usage export can help identify usage by API key and model. (Anthropic’s model deprecations page)
- Test Sonnet 5.5 on representative work. Include normal and difficult requests, tool-use flows, structured outputs, error handling, and latency or cost requirements. Anthropic recommends testing replacement models on your own tasks before retirement. (Anthropic’s model deprecations page)
- Review request compatibility; do not just swap the model name. Anthropic’s Sonnet 5.5 migration guide identifies request settings that can produce a 400 error, including sampling parameters such as
temperature,top_p, andtop_k, as well as changes involving thinking, assistant prefills, and forced tool choice. Check the guide for the exact changes relevant to your Messages API requests. - Compare outputs and cost, then switch before the deadline. A replacement recommendation does not establish behavioral or cost equivalence for your application. Validate your acceptance criteria and update production configuration before requests to a retired model fail. (Anthropic’s model deprecations page)
If you use Amazon Bedrock, follow AWS’s lifecycle information instead of applying Anthropic API dates automatically. AWS lists Sonnet 4.5’s Bedrock end-of-life date as April 8, 2027. (AWS Bedrock model card)
Sources
Claude Sonnet 4.5 — Claude Platform Docs
Model deprecations — Claude Platform Docs
Pricing — Claude Platform Docs
Claude Sonnet 5.5 — Claude Platform Docs
Claude Sonnet 5.5 migration guide — Claude Platform Docs
Introducing Claude Sonnet 4.5 — Anthropic