Anthropic shipped Claude Sonnet 4.6 today, November 8, 2026, and the headline number is the one everyone will quote: a 1-million-token context window, now in beta. But after reading the company's official announcement and cross-checking it against same-day coverage, I think the real story is quieter — the mid-tier model just got good enough that most people will stop reaching for the expensive one.
How I checked this: I read Anthropic's official announcement, "Introducing Sonnet 4.6," published November 8, 2026, and checked it against DEV Community's same-day breakdown of the launch. Every number below is Anthropic's own claim from its early testing — I have not run Sonnet 4.6 hands-on, so treat this as a launch-day read, not a review.
What actually changed
Anthropic calls Sonnet 4.6 its most capable Sonnet model yet — a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. It is now the default model in claude.ai and Claude Cowork on the Free and Pro plans, and it is available on all Claude plans, Claude Code, the API, and all major cloud platforms. The API model ID is claude-sonnet-4-6.
A million tokens — but can it think across them?
A 1M-token window can hold an entire codebase, a long contract, or dozens of research papers in a single request. The size alone isn't the story; Anthropic's claim is that Sonnet 4.6 reasons effectively across all of it. Its evidence is a long-horizon test called Vending-Bench Arena, where the model runs a simulated business over time — Anthropic says Sonnet 4.6 invested in capacity early, then pivoted to profitability, finishing well ahead of the competition. The window is in beta, and Anthropic also added context compaction in beta, which automatically summarises older context as conversations approach their limits.
Coders preferred it over the frontier model
Here is the number that made me sit up. In Anthropic's early testing inside Claude Code, users preferred Sonnet 4.6 over Sonnet 4.5 roughly 70% of the time — and preferred it to Opus 4.5, the frontier model from November 2025, 59% of the time. Users reported it read the context before modifying code, consolidated shared logic instead of duplicating it, and was less prone to overengineering, with better instruction following and more consistent follow-through on multi-step tasks. Again: Anthropic's own testing, not mine. But if even half of that holds up in real repositories, daily coding changes more than any benchmark suggests. Coding assistants are racing on the same ground — GitHub Copilot Fusion just went generally available with the same pitch of cheaper, smarter AI coding — so this preference claim will get tested in public very fast.
Computer use is finally useful — with a catch
On OSWorld, the standard benchmark for AI computer use, Sonnet 4.6 scored 72.5%, up from 61.4% for Sonnet 4.5. Anthropic says early users are seeing human-level capability in tasks like navigating a complex spreadsheet or filling out a multi-step web form across several browser tabs — while admitting the model still lags behind the most skilled humans. Just as important: prompt-injection resistance is a major improvement over Sonnet 4.5 and now performs similarly to Opus 4.6. If you let a model click around on your behalf, that second sentence matters at least as much as the first.
The price didn't move — and the free tier got real
Pricing is unchanged from Sonnet 4.5: starting at $3 per million input tokens and $15 per million output tokens. No price hike, no new tier. The free tier got a genuine upgrade too: it now includes file creation, connectors, skills, and compaction, with Sonnet 4.6 as the default. If you've never tried Claude, our beginner's guide to using Claude in 2026 walks through the setup step by step — the free tier excuse just expired.
Why this matters
This is the moment the mid-tier model stops being the compromise. Anthropic is unusually clear about the split: use Sonnet 4.6 for the overwhelming majority of daily work, and keep Opus 4.6 for the deepest reasoning — large refactors, coordinating multiple agents, problems where getting it just right matters most. The practical rule is simple: start on Sonnet, escalate only the tasks that smell like a reasoning ceiling. For Anthropic, the business logic is just as clear — a cheaper model that does frontier work keeps more users inside its ecosystem, and the free-tier upgrade is a direct invitation to people who never tried Claude Code or Cowork. It also continues a pattern: Anthropic cut Haiku 5.5's price by 90% last month, and now the middle of the lineup got stronger without getting dearer. The race is shifting from "who has the smartest model" to "whose good-enough model costs least."
The limits, stated plainly
Launch-day numbers are promises, not reviews — the 70% and 59% preference figures come from Anthropic's early testing with its own users, and independent benchmarks will tell us how much survives contact with other people's code. The 1M window is beta, so expect rough edges, and more context costs more per request — every token you send is a token you pay for. Opus still wins at the top end by Anthropic's own account. And computer use is much better, not human — giving a model control of your screen still deserves caution, even with better prompt-injection resistance. Meanwhile, OpenAI's newest model just took a very different kind of exam — a wet-lab benchmark where GPT-6 Astra scored 66.7% — a reminder that the frontier race now runs on many tracks at once.
Verdict
Sonnet 4.6 is not a revolution. It is something more useful: the moment the mid-tier model got good enough that the expensive model becomes a specialist tool instead of a default. Same price, a million-token window in beta, coding users apparently preferring it over the previous frontier model, and a free tier that is genuinely usable. If you write code for a living, the move is simple: switch your default to claude-sonnet-4-6, keep Opus in reserve, and watch what changes in your own workflow over the next week. Then tell us — did it actually read the context before editing, or did it still duplicate the helper function?
Sources
- Anthropic — "Introducing Sonnet 4.6", published November 8, 2026
- DEV Community — "Claude Sonnet 4.6 Is Here: 1M Context, Opus-Level Coding at Sonnet Prices", checked November 8, 2026


Comments
Post a Comment