Overview
Claude Sonnet 4.5 was released on September 29, 2025, and at launch posted the highest SWE-bench Verified score of any Claude model to date: 77.2% (rising to 82.0% with parallel test-time compute). It also set a record on OSWorld for computer use at 61.4%, well above the 42.2% Sonnet 4 had managed. Anthropic positioned it as a drop-in replacement for Sonnet 4 at the same price.
Sonnet 4.5 is a hybrid reasoning model with extended thinking, and it was the model that introduced sustained long-horizon autonomy: Anthropic reported it could stay on task for more than 30 hours on complex multi-step work. Alongside the model, Anthropic shipped context editing, a memory tool, checkpoints, and a VS Code integration to support these long-running agentic workflows.
Sonnet 4.5 carries a 200K token context window (with a 1M token beta), 64K max output tokens, and supports text and vision input. It remains available on the Claude API as claude-sonnet-4-5, on Amazon Bedrock, and on Vertex AI, and is priced at $3 / $15 per million tokens.
| Released | 2025-09-29 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Context | 1M |
| Max output | 64K |
| Architecture | Proprietary transformer; hybrid reasoning model with extended thinking and effort control. |
| Knowledge cutoff | January 2025 |
| Modalities | Text, Vision, PDF |
| Status | Generally available |
Benchmarks
- SWE-bench Verified77.2%
- OSWorld (computer use)61.4%
Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.
Pricing
| Input | $3.00 per million tokens |
|---|---|
| Output | $15.00 per million tokens |
Same price as Sonnet 4; one-fifth the price of Opus 4.1.
Strengths
- Record SWE-bench Verified (77.2%) and OSWorld (61.4%) scores at launch
- Sustained autonomous agent operation for 30+ hours on complex tasks
- Extended thinking for multi-step reasoning in math, science and coding
- Shipped with context editing, memory tool, and checkpoints for agents
- Same $3/$15 pricing as Sonnet 4, a fifth of Opus 4.1's cost
Best for
- Long-running autonomous coding agents and multi-hour build tasks
- Computer-use automation across desktop and browser interfaces
- Agentic workflows needing memory and context-editing tooling
- Domain-specific knowledge work in finance, law, medicine and STEM
- IDE-integrated coding assistance (VS Code, Claude Code)
How to access
| Provider | Model ID |
|---|---|
| Anthropic ↗ | claude-sonnet-4-5 |
| Amazon Bedrock ↗ | anthropic.claude-sonnet-4-5-20250929-v1:0 |
| Google Vertex AI ↗ | claude-sonnet-4-5@20250929 |
Claude Sonnet — every version
The full lineage of the Claude Sonnet line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Claude Sonnet 4.6current | 2026-02-17 | 1M | Proprietary |
| Claude Sonnet 4.5 | 2025-09-29 | 1M | Proprietary |
| Claude Sonnet 4 | 2025-05-22 | 200K | Proprietary |
| Claude 3.7 Sonnet | 2025-02-24 | 200K | Proprietary |
| Claude 3.5 Sonnet (new) | 2024-10-22 | 200K | Proprietary |
| Claude 3.5 Sonnet | 2024-06-20 | 200K | Proprietary |
| Claude 3 Sonnet | 2024-03-04 | 200K | Proprietary |
FAQ
What was special about Claude Sonnet 4.5?
At its September 2025 launch it had the best SWE-bench Verified score of any Claude model (77.2%) and the best OSWorld computer-use score (61.4%), plus the ability to run agents autonomously for over 30 hours.
How much does Claude Sonnet 4.5 cost?
$3 per million input tokens and $15 per million output tokens, the same as Sonnet 4 and one-fifth the price of Opus 4.1.
What is the context window?
200K tokens standard, with a 1 million token beta context window and up to 64K output tokens.
Is Sonnet 4.5 still available?
Yes. It remains active on the Claude API as claude-sonnet-4-5, and on Amazon Bedrock and Vertex AI, though Anthropic recommends Sonnet 4.6 for new work.