Direct answer
Claude Sonnet 4.5 is Anthropic’s Sonnet model released September 29, 2025. The direct Claude API identifies it with the dated model ID claude-sonnet-4-5-20250929; claude-sonnet-4-5 is also a listed API alias. As of November 9, 2026, Anthropic lists the model as deprecated: it still works but is no longer recommended, and is scheduled to retire from Anthropic-operated platforms on November 30, 2026. Anthropic recommends Claude Sonnet 5.5 as the replacement. For new production work, evaluate an active model; for an existing Sonnet 4.5 integration, plan a tested migration before the retirement date that applies to your provider. [1]
Model identity and availability
Anthropic’s model reference lists a 200,000-token context window, 64,000-token maximum output, and January 2025 reliable knowledge cutoff for Sonnet 4.5. These are model-reference specifications; verify applicable limits and availability for the provider route you use. [1]
| Route | Model identifier |
|---|---|
| Claude API, dated ID | claude-sonnet-4-5-20250929 |
| Claude API alias | claude-sonnet-4-5 |
| Amazon Bedrock InvokeModel | anthropic.claude-sonnet-4-5-20250929-v1:0 |
The identifiers are route-specific, not interchangeable universal names. Anthropic-operated and partner-operated platforms can also have different lifecycle schedules. Anthropic lists November 30, 2026 as its retirement date for this model; AWS lists a separate April 8, 2027 end-of-life date for Bedrock. Follow the schedule for the platform you actually call. [1][2]
Workload fit and evidence
Anthropic’s launch announcement positioned Sonnet 4.5 for coding, complex agents, and computer use. It reported a 61.4% OSWorld result, compared with Sonnet 4’s earlier 42.2%, and said the model maintained focus for more than 30 hours on complex, multi-step tasks. Treat these as Anthropic’s launch-era results and claims—not guarantees for your application or a current comparison with active models. [3]
Use those claims to decide what to evaluate, not as a substitute for testing:
- Coding: Run representative repository tasks and measure completion, correctness, review effort, latency, and cost.
- Tool-using agents: Test the full loop, including tool selection, permissions, error recovery, and human approval. Claude returns a structured tool request; your application or an Anthropic-hosted tool runs the operation. [4]
- Computer use: Anthropic lists Sonnet 4.5 as a supported model for computer use. The available reference text does not explicitly map a particular earlier tool version and beta-header combination to Sonnet 4.5 on the direct Claude API. Confirm the compatible version and required header in the current tool reference for your route. The tool runs in an environment controlled by your application. [5]
For a new deployment, compare an active model on your own task set. Sonnet 4.5’s historical benchmark results do not outweigh its approaching first-party API retirement. Anthropic recommends Sonnet 5.5; confirm replacement quality and tool behavior with your workload before migrating. [1]
Production API basics
For direct Anthropic access, conversational requests use the Messages API at POST /v1/messages. A basic request includes model, max_tokens, and messages. Authenticate with an API key or configured Workload Identity Federation; the API accepts bearer-token authorization, and its overview documents x-api-key as an alternative authentication header. The Messages reference labels its current API version 2026-10-21; consult the API overview for the required request headers for your route rather than treating that version label alone as a complete header specification. Add an anthropic-beta header when a beta feature requires one. [6][7]
The response is an assistant message with a content array of typed blocks, such as text or tool_use, and a stop_reason. For a client-side tool call, your code runs the requested operation and sends a new Messages request with a user message containing a tool_result block whose tool_use_id matches the original call. Claude can then continue the conversation. The API is stateless by default, so send the relevant conversation history with each request. [4][8][10]
Before production use, verify the exact route and ID, lifecycle date, model and feature limits, and any tool-version or beta-header requirements. Test the chosen model and its replacement on representative inputs, checking output quality, structured results, tool-call behavior, latency, and spend. Partner APIs may have different identifiers and lifecycle dates; do not assume Anthropic’s direct-API details automatically apply to another host. [1][2][5]
API price reference
Anthropic’s pricing page lists Sonnet 4.5 at US$3 per million input tokens and US$15 per million output tokens for the Claude API. At those base rates, one million input tokens plus one million output tokens would cost US$18, before applicable caching, batch, or other pricing options. This is API usage pricing, not a Claude app subscription price; partner platforms may bill under their own terms. Given the approaching first-party retirement, treat this as a reference for existing workloads, not a reason to start a new one on Sonnet 4.5. [9]
Bottom line
Use an active model for a new production integration. Sonnet 4.5’s documented API supports tool use and computer-use workflows, but the model is deprecated and scheduled to retire from Anthropic-operated platforms on November 30, 2026. If you already use it, inventory the routes that depend on it, test the recommended replacement, and complete migration according to the retirement date for each provider. [1][2][5]
Sources
Claude Sonnet 4.5 — Claude Platform Docs
Claude Sonnet 4.5 — Amazon Bedrock Documentation
Introducing Claude Sonnet 4.5 — Anthropic
Tool use with Claude — Claude Platform Docs
Computer use tool — Claude Platform Docs
API overview — Claude Platform Docs
Authentication — Claude Platform Docs
Using the Messages API — Claude Platform Docs
Pricing — Claude Platform Docs