Claude Sonnet 4.6 Is Here: 1M Context, Opus-Level Coding at Sonnet Prices

Nothing here is sponsored — no affiliate links on this page. This breaking article is not the daily buyer-intent post. Facts below come from the sources listed at the end, checked 8 November 2026. We have not tested Sonnet 4.6 hands-on, and nothing here is a hands-on finding.

Anthropic shipped Claude Sonnet 4.6 today, 8 November 2026, and it is the most interesting Sonnet release in a while — not because of one flashy feature, but because of what it quietly makes unnecessary. For a lot of everyday coding work, you no longer need to reach for the most expensive model in the lineup. Sonnet now does work that used to require Opus, at Sonnet prices.

I have not used Sonnet 4.6 hands-on, and I am not going to pretend otherwise. Everything below is what Anthropic published in its announcement, checked 8 November 2026. The numbers in this article — the 70%, the 59%, the benchmark gains — are Anthropic's own claims from its early testing, not independent measurements. I will say so where it matters, because a launch-day announcement is a promise, not a review.

What actually changed

Sonnet 4.6 is, in Anthropic's words, its most capable Sonnet model yet: a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. Three things in that list are worth pulling out.

The 1M token context window (in beta). A million tokens is enough to hold entire codebases, lengthy contracts, or dozens of research papers in a single request. But the more important claim is not the size — it is that Sonnet 4.6 reasons effectively across all that context. A big window you cannot think across is just a bigger place to get lost. Anthropic points to its Vending-Bench Arena result as evidence: in that long-horizon test, where models run a simulated business over time, Sonnet 4.6 invested heavily in capacity for the first ten simulated months, then pivoted sharply to profitability in the final stretch — and finished well ahead of the competition. That is a planning story, not a memory story, and it is the one to watch.

Coding that users preferred over Opus. In Claude Code, Anthropic's early testing found 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 significantly less prone to overengineering and "laziness", with better instruction following, fewer false claims of success, 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, the day-to-day coding experience changes more than any benchmark number suggests.

Computer use grew up. On OSWorld, the standard benchmark for AI computer use — hundreds of tasks across real software like Chrome, LibreOffice, and VS Code, with no special APIs, just a model clicking and typing like a person — Sonnet models have made steady gains over sixteen months, and 4.6 is a major step. Anthropic says early users are seeing human-level capability in tasks like navigating a complex spreadsheet or filling out a multi-step web form, then pulling it all together across multiple browser tabs. The model still lags behind the most skilled humans, Anthropic admits. But computer use is now useful for a range of real work tasks, not just demos. 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.

What 1M context actually means for you

A million tokens sounds abstract. In practice, it means you can stop doing the thing every developer hates: chopping a codebase into pieces small enough to fit, then losing the thread between the pieces.

• Whole-repo refactors in one conversation. Load the service, its tests, and its config, and ask for a change that touches all three — with the model actually holding all three in mind.
• Long documents, read properly. A hundred-page contract or a stack of research papers, questioned as one body of text instead of in slices.
• Agents that plan further ahead. Long-horizon work — a multi-step migration, a business simulation, a research project — benefits most, because the model can keep the plan and the evidence in the same head.

Two honest caveats. First, the 1M window is in beta, and beta means the edges are still being found. Second, a bigger window is not a reason to dump everything into every prompt. Context still costs money and attention — every token you send is a token you pay for, and a model reasoning across a million tokens is doing more work than one reasoning across ten thousand. The skill that matters now is not "fit more in" but "put the right things in". Anthropic's context compaction feature, also in beta, helps here: it automatically summarises older context as conversations approach limits, increasing the effective length. Use it, and curate what goes in.

Sonnet 4.6 vs Opus 4.6: when to use which

This is the question that actually decides your bill, so let us be precise. Anthropic is unusually clear about it:

Use Sonnet 4.6 for the overwhelming majority of daily work — writing and reviewing code, agent planning, long-context reasoning, knowledge work, design, and computer use. It now handles performance that previously required an Opus-class model, including on real-world, economically valuable office tasks. At $3/$15 per million tokens — the same price as Sonnet 4.5 — it is the default choice, and it is now the default model in claude.ai and Claude Cowork on Free and Pro plans.

Use Opus 4.6 for the tasks that demand the deepest reasoning: large codebase refactoring, coordinating multiple agents in a workflow, and problems where getting it just right is paramount. Opus 4.6 remains the strongest option there. Anthropic says so plainly, and you should believe them — the whole point of Sonnet 4.6 is that you reach for Opus less often, not never.

The practical rule: start on Sonnet 4.6. If a task comes back shallow, inconsistent, or wrong in a way that smells like a reasoning ceiling rather than a prompt problem, escalate that one task to Opus. Do not default everything to the most expensive model out of habit — that habit just got a lot more expensive to justify.

Pricing and how to get it

Pricing is unchanged from Sonnet 4.5: starting at $3/$15 per million tokens. No price hike, no new tier to decode.

Sonnet 4.6 is available now on all Claude plans, Claude Cowork, Claude Code, the API, and all major cloud platforms. The API model ID is claude-sonnet-4-6. The free tier got a real upgrade alongside it: it now includes file creation, connectors, skills, and compaction, with Sonnet 4.6 as the default. If you have never tried Claude Code or Cowork because the free tier felt too thin, that excuse just expired.

On the platform side, Sonnet 4.6 supports adaptive thinking and extended thinking, and the API's web search and fetch tools now automatically write and execute code to filter and process search results — keeping only relevant content in context, which should improve both response quality and token efficiency. Code execution, memory, programmatic tool calling, tool search, and tool use examples are now generally available. And for Excel users, the Claude add-in now supports MCP connectors — S&P Global, LSEG, Daloopa, PitchBook, Moody's, FactSet — on Pro, Max, Team, and Enterprise plans.

The limits, stated plainly

• No hands-on testing yet. This article is based on Anthropic's announcement. The preference numbers (70% over Sonnet 4.5, 59% over Opus 4.5) come from Anthropic's early testing with its own users. Independent benchmarks and real-world reports will tell us how much of this survives contact with other people's codebases.
• 1M context is beta. Expect rough edges, and expect the effective behaviour to differ from the headline number. Also remember: more context costs more per request.
• Opus still wins at the top end. Deepest reasoning, multi-agent coordination, highest-stakes correctness — that is still Opus 4.6's job, by Anthropic's own account.
• Computer use is much better, not human. Anthropic says so itself. And giving a model control of your screen still deserves caution, even with better prompt-injection resistance.
• Design gains are customer-reported. Early customers described frontend output as more polished, with better layouts, animations, and design sensibility, and fewer rounds of iteration to reach production quality. Encouraging — and exactly the kind of claim that needs a week of real usage to confirm.

The bottom line

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 as Sonnet 4.5, a million-token context window in beta, coding users apparently prefer over the previous frontier model, computer use that finally works on real tasks, 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 today, keep Opus 4.6 in reserve for the hardest problems, and watch what changes in your own workflow over the next week. Then tell me — did it actually read the context before editing, or did it still duplicate the helper function? I am genuinely curious, because that one behaviour is the difference between a model upgrade and a workflow upgrade.

Sources: Anthropic — "Introducing Sonnet 4.6" (anthropic.com/news/claude-sonnet-4-6), checked 8 November 2026. All facts, figures and preference percentages are Anthropic's claims from its announcement and early testing, stated as such. No hands-on testing was performed for this article. The 1M token context window is in beta. Opus 4.6 remains Anthropic's strongest option for the deepest reasoning tasks. This breaking article is not the daily buyer-intent post, and it contains no affiliate links.