Claude Sonnet 4.6 is here — Anthropic's smartest Sonnet yet

Anthropic launched Claude Sonnet 4.6 on Feb 17, 2026. It's their most capable Sonnet model so far — a full step up in coding, computer use, long-context reasoning, agent planning, knowledge work, and design. Sonnet 4.6 also ships with a 1M token context window in beta, so you can throw in whole codebases or stacks of docs and have it reason across them.

What it's for
If you've been reaching for Opus for heavy lifting, 4.6 is meant to close that gap at Sonnet pricing. Early users preferred it over Sonnet 4.5 about 70% of the time in Claude Code, and even chose it over Opus 4.5 roughly 59% of the time — less overengineering, better instruction following, fewer fake "done!" moments. So you get Opus-style performance on a lot of real work without the higher cost.

Why it matters

Computer use: Sonnet has been getting steadily better at using a computer like a human — click, type, switch tabs. On OSWorld (the standard benchmark for that), 4.6 keeps the upward trend. People are already using it for things like complex spreadsheets and multi-step forms. Anthropic also tightened prompt injection resistance, so it's safer when reading sketchy web content.
Context that actually works: The 1M-token window isn't just big — the model reasons across it. In their Vending-Bench Arena eval (simulated business over time), 4.6 tried a different strategy: invest in capacity early, then pivot to profitability late, and came out ahead. That kind of long-horizon planning is where the bigger context really pays off.
Design and frontend: Several customers called out more polished layouts, animations, and design sense — and fewer rounds of tweaks to get to something shippable.

Where you get it

Everyone: claude.ai and Claude Cowork — Sonnet 4.6 is now the default on Free and Pro. Pricing stays the same as 4.5 ($3 / $15 per million tokens). The free tier now includes file creation, connectors, skills, and context compaction.
Developers: Use claude-sonnet-4-6 on the Claude API; adaptive thinking, extended thinking, and context compaction (beta) are supported. Web search and fetch now auto-filter results with code for better quality and token use.
Excel users: Claude in Excel supports MCP connectors (S&P Global, LSEG, PitchBook, etc.) on Pro, Max, Team, and Enterprise.

Bottom line: if you wanted Opus-level smarts for coding, docs, or computer use but didn't want to pay Opus prices, Sonnet 4.6 is the one to try. Same safety bar as recent Claudes, and a clear step up from 4.5.

Try it: Open claude.ai or your usual Claude app — 4.6 is already the default. For the API, call claude-sonnet-4-6 and you're set.