Anthropic's most capable Sonnet-class model. 79.6% SWE-bench coding, 91.7% τ-bench tool use, and 1M token context — frontier performance at everyday prices.
No credit card required
Frontier intelligence without the frontier price tag
Near Opus-level τ-bench scores (91.7%) — Sonnet perfectly coordinates terminal commands, file edits, and web searches.
At $3/$15 per 1M tokens, Sonnet delivers 79.6% SWE-bench coding performance at 5x less cost than Opus.
Extended thinking, adaptive thinking, and configurable effort levels let Sonnet reason deeply on hard problems while staying fast on simple ones.
Process entire codebases, long documentation sets, and complex multi-file architectures in a single pass.
79.6% on SWE-bench Verified — within 1.2 points of Opus 4.6, the industry leader, at a fraction of the price.
Significantly faster generation than Opus, making it the ideal daily driver for iterative development workflows.
Near-Opus intelligence at a fraction of the cost
| Benchmark | Sonnet 4.6 | Opus 4.6 | Gemini 3.1 Pro | GPT-5.2 |
|---|---|---|---|---|
| SWE-bench Verified (Agentic coding) | 79.6% | 80.8% | 80.6% | 80.0% |
| Terminal-Bench 2.0 (Agentic terminal coding) | 59.1% | 65.4% | 68.5% | 64.7% |
| BrowseComp (Agentic search) | 74.7% | 84.0% | 85.9% | 65.8% |
| τ-bench Retail (Agentic tool use) | 91.7% | 91.9% | 90.8% | 82.0% |
| ARC-AGI-2 (Abstract reasoning) | 58.3% | 68.8% | 77.1% | 52.9% |
Source: Anthropic — Introducing Claude Sonnet 4.6 · Google DeepMind Model Card
Get started in minutes
Install globally with npm. Works on macOS, Linux, and Windows.
Create your account — takes less than a minute.
Use /model to pick Claude Sonnet 4.6 from the model selector.
Experience near-Opus coding intelligence with Sonnet-class speed and pricing in AdaL CLI.
Get Started with AdaL