onllm.dev
What Changed
Anthropic released Claude Sonnet 4.6 on February 17, 2026, two weeks after Opus 4.6 and consistent with the company's four-month release cadence. The mid-tier model doubles its context window to 1 million tokens (in beta) and shows improvements across coding and instruction-following tasks.
Sonnet 4.6 becomes the default model on Free and Pro plans in claude.ai and Claude Cowork. The model maintains the same pricing as its predecessor: $3 per million input tokens and $15 per million output tokens.
In internal testing with Claude Code, users preferred Sonnet 4.6 over its predecessor 70% of the time. More notably, users preferred Sonnet 4.6 over Opus 4.5 (released November 2025) 59% of the time -- a mid-tier model outperforming the previous flagship in user preference.
Context and Coding Improvements
The 1 million token context window handles entire codebases, lengthy contracts, or dozens of research papers in a single conversation. This capacity requires the context-1m-2025-08-07 header and remains in beta. Standard context remains 200K tokens.
Anthropic reports specific coding improvements: better context reading, less code duplication, and less overengineering. The model also shows fewer hallucinations and fewer false claims of task completion.
Maximum output increased to 64K tokens. The model supports both extended thinking and adaptive thinking modes.
Training data extends to January 2026, with reliable knowledge through August 2025.
Benchmarks
Sonnet 4.6 scores 60.4% on ARC-AGI-2, the reasoning benchmark. This trails higher-tier models -- Opus 4.6, Gemini 3 Deep Think, and refined GPT 5.2 all score higher.
On coding benchmarks, Sonnet 4.6 sets records on SWE-Bench, the software engineering evaluation suite. The model also achieves a record score on OSWorld, the computer use benchmark. Early users report human-level capability on computer use tasks.
Anthropic's safety evaluation found "very strong safety behaviors, and no signs of major concerns."
Pricing and Availability
| Provider | Model ID | Context |
|---|---|---|
| Anthropic API | claude-sonnet-4-6 |
200K / 1M beta |
| AWS Bedrock | anthropic.claude-sonnet-4-6 |
200K / 1M beta |
| GCP Vertex AI | claude-sonnet-4-6 |
200K / 1M beta |
Pricing remains unchanged from Sonnet 4.5:
| Model | Input | Output |
|---|---|---|
| Sonnet 4.6 | $3/M tokens | $15/M tokens |
| Opus 4.6 | $5/M tokens | $25/M tokens |
Sonnet 4.6 costs 40% less than Opus 4.6 on input and output.
An updated Haiku model is expected in coming weeks.