Anthropic releases Claude Sonnet 4.6: Benchmark performance, how to try it
Source: Mashable Tech
Anthropic has just released its latest large language model, Claude Sonnet 4.6, following the launch of Claude Opus 4.6 on Feb. 5. According to the company, Sonnet 4.6 is “our most capable Sonnet model yet” and offers a 1 million‑token context window (currently in beta). Internal safety tests show a low tendency to hallucinate or engage in sycophancy, and the model’s coding abilities have been notably improved for developers.
How to access Claude Sonnet 4.6
- Web UI: Available as the default model on claude.ai and Claude Cowork for both free and Pro users.
- API & Cloud: Rolled out through Anthropic’s API and major cloud platforms.
Pricing
| Plan | Cost |
|---|---|
| Free tier | Limited usage rates that reset every 5 hours (subject to demand). |
| Claude Pro | $20 / month (or $17 / month when billed annually). |
| API usage | $3 per million input tokens and $15 per million output tokens. |
| Claude Opus 4.6 (for comparison) | $5 per million input tokens and $25 per million output tokens. |
Claude Sonnet 4.6 benchmark performance
Anthropic’s internal benchmarks show Sonnet 4.6 as the most powerful model for agentic financial analysis and office tasks, outperforming competitors such as Google’s Gemini 3 Pro and OpenAI’s GPT 5.2, as well as Anthropic’s own Opus 4.6 on those specific tasks. Early‑access developers preferred Sonnet 4.6 over both its predecessor (Sonnet 4.5) and Opus 4.5.
Benchmark results
- GPQA Diamond: 89.9 %
- ARC‑AGI‑2: 58.3 %
- MMMLU: 89.3 %
- SWE‑bench Verified: 79.6 %
- Humanity’s Last Exam (HLE):
- With tools: 49.0 %
- Without tools: 33.2 %
AI‑powered insurance company Pace reported that Sonnet 4.6 achieved the highest score among Claude models on its complex insurance computer‑use benchmark.
These results highlight that while Claude Opus models generally excel at complex reasoning, Sonnet 4.6 offers a strong combination of performance and affordability.