AnthropicGA
Claude Sonnet 5
The new mid-tier default — 1M context and frontier computer-use at Sonnet economics.
Context
1M
Max output
64K
Input /1M
$2.00
Output /1M
$10.00
Live pricing via OpenRouter
Best for
- Production coding and integrations at $3/$15
- Computer-use agents (81.2% OSWorld, vendor-reported)
- 1M-context work without flagship pricing
Watch out
Opus still leads hard agentic coding — 69.2% vs 63.2% SWE-Bench Pro, which Anthropic states plainly. Launch benchmarks are vendor-reported. The $2/$10 introductory rate ends 2026-08-31; budget against $3/$15.
For creators. The reliable default for content-engine and production coding work. Route to Opus only when the task earns the premium.
Benchmarks
| swe bench pro | 63.2 |
| osworld | 81.2 |
| gpqa diamond | 91.1 |
| aa intelligence index | 55.3 |
Capabilities
- 1M context window
- Computer use - 81.2% OSWorld (vendor-reported)
- Extended reasoning, tool calling, prompt caching, structured outputs
- Default Claude model for Free and Pro on claude.ai; available in Claude Code, the API, Cursor, VS Code and GitHub Copilot
- Introductory $2/$10 pricing through 2026-08-31