Sonnet 4.6 became the default for free and paid users at unchanged pricing, with a million-token context window in beta.
Default-model changes matter more than flagship launches, because they reach reps who never chose a model. If your team uses Claude at all, the thing answering them changed without anyone opting in — and the context window now plausibly fits a full account history in one pass.
A context window is how much text a model can hold in view at once. Bigger window, more source material per answer without splitting it into chunks. One million tokens is roughly a few thousand pages.
Anthropic positioned 4.6 as its most capable Sonnet, with gains in long-context reasoning, agent planning and knowledge work. Pricing held at the previous tier, which is the unusual part — capability normally arrives with a price change.
The short version: the model most people were already using got better and cheaper per unit of work, and it can now read far more at once.
Anthropic released Claude Sonnet 4.6 on 17 February 2026, describing it as its most capable Sonnet model, with improvements across coding, computer use, long-context reasoning, agent planning, knowledge work and design.
Availability: Default model for Free and Pro users on claude.ai and Claude Cowork; available across all Claude plans, Claude Code, the API and major cloud platforms.
Price: Unchanged from Sonnet 4.5, starting at $3 per million input tokens and $15 per million output tokens.
Context window: One million tokens, in beta.
Regions: Not stated.
Migration: Not stated — no deprecation date given for the previous default.
What it doesn't do: Anthropic states plainly that the model still lags skilled humans at using computers, and that computer use carries prompt-injection risk. It also recommends Opus 4.6 over Sonnet for the deepest reasoning work, so this is not the top of the range. The million-token window is beta, not a committed capability.
SDR — Little direct change to daily drafting — model quality was not your constraint. The free-tier default upgrade does mean reps without a paid seat are no longer working with materially worse output.
AE — The long-context gain is the story. Whole-deal analysis — every call transcript, the RFP, the email thread, assessed together rather than summarised piecemeal — becomes practical rather than theoretical.
Manager — Quarter-over-quarter output quality moved without any change in process, which confounds coaching signals. If rep output improved this quarter, some of that is the model, not the coaching.
RevOps — Cost planning is the relevant angle: unchanged token pricing with better capability means your existing API budget stretches further, and the 1M window is beta rather than committed.
A bigger context window is only worth having if you ask it a question worth the tokens. Put the whole quarter in and make it argue three cases rather than one.
Two minutes, once a week. What changed in AI, and what to run because of it.