News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgfnClaude Sonnet 4.6

Claude Sonnet 4.6

Anthropic's Sonnet-tier model for coding, computer use and long-context work

Claude Sonnet 4.6 is a general-purpose language model from Anthropic, positioned as the strongest Sonnet-tier release to date. It handles coding, computer use, long-context reasoning, agent planning and knowledge work, with a 1M token context window offered in beta.

Share on WhatsApp

  • 1M token context window available in beta
  • Pricing unchanged from Sonnet 4.5: $3/$15 per million tokens
  • Default model for Free and Pro plans on claude.ai and Claude Cowork
  • Users preferred it to Opus 4.5 59% of the time in Claude Code testing
  • Anthropic reports better prompt injection resistance than Sonnet 4.5

What it is

Claude Sonnet 4.6 is a language model built by Anthropic and offered through the company's own products. Anthropic presents it as the most capable Sonnet-class model so far, covering coding, computer use, long-context reasoning, agent planning, knowledge work and design. It accepts a context window of one million tokens in beta, enough for whole codebases or long document sets in one request. Pricing is unchanged from Sonnet 4.5, starting at $3 per million input tokens and $15 per million output tokens. On claude.ai and Claude Cowork it is the default for Free and Pro users.

How it came about

Anthropic's computer-use work began in October 2024, when the company says it shipped the first general-purpose model able to operate software, describing it at the time as experimental and error-prone. Sonnet 4.6 follows Sonnet 4.5 and sits below Opus 4.5, the company's November 2025 frontier model. Developers given early access reportedly favoured the new release over its predecessor by a wide margin, and often over Opus 4.5 as well. A system card covering capabilities and safety behaviour accompanies the launch.

Inside the model

Anthropic has not published layer counts, parameter totals or training data for this model. The company describes it as a full upgrade across coding, computer use, long-context reasoning, planning and knowledge work. Its one-million-token window lets it hold large codebases, contracts or research collections at once, and the company says reasoning stays effective across that span. For computer use, the model reads a screen and acts through a virtual mouse and keyboard, with no special connectors. Safety testing covered prompt injection resistance, where the company reports a clear gain over Sonnet 4.5 and performance close to Opus 4.6.

What it can do

Anthropic reports gains on benchmarks across the board and says the model approaches Opus-level capability at Sonnet pricing. In Claude Code testing, users picked it over Sonnet 4.5 about 70% of the time and over Opus 4.5 59% of the time, citing fewer invented successes and better instruction following. On OSWorld, a computer-use benchmark of hundreds of tasks in real applications, Sonnet models have improved steadily over sixteen months; scores before Sonnet 4.5 used the original OSWorld, later ones OSWorld-Verified. Customers cited strong frontend output, document comprehension and bug detection. The company still notes a gap versus the best human operators.

How to use it

The model is served through claude.ai and Claude Cowork, where it is the default for Free and Pro plans, and through the API. Token pricing matches Sonnet 4.5 at $3 per million input and $15 per million output. A one-million-token context window is available in beta. Guidance on prompt injection and other risks appears in Anthropic's API documentation.

Safety and testing

Anthropic says it ran extensive safety evaluations and judged the model as safe as, or safer than, its other recent Claude models. Its researchers described a warm, honest, prosocial character with strong safety behaviour and no major misalignment concerns in high-stakes areas. Computer use brings prompt injection risk, where hidden instructions on websites try to seize control; the company reports better resistance than Sonnet 4.5 and similar results to Opus 4.6.

Reception and debate

Early users and customers reported broad improvements, with frontend code and financial analysis singled out. Several firms described more polished visual output, fewer revision rounds before production quality, and better results on long-horizon or branched tasks. One insurance benchmark reached 94%, which a customer called the best computer-use result they had tested. Others noted a narrower gap with Opus on bug detection and stronger retrieval in financial workflows. Box measured a 15 percentage point gain over Sonnet 4.5 on heavy reasoning questions over enterprise documents.

Why it matters

Anthropic frames the release as bringing performance that once needed an Opus-class model into a cheaper tier, widening access for everyday coding and office tasks. The company's computer-use line, running since October 2024, has moved from clumsy and error-prone to what early users describe as human-level on tasks such as spreadsheet navigation and multi-step web forms. Anthropic says the pace of progress suggests more capable models are close, while acknowledging the model still trails expert humans at operating computers.

Questions people ask about Claude Sonnet 4.6

What is Claude Sonnet 4.6 for?

Anthropic positions it as a general-purpose model aimed at coding, computer use, long-context reasoning, agent planning and knowledge work. It is meant to handle tasks that previously called for a more expensive Opus-class model, at Sonnet pricing.

How much does it cost?

Anthropic kept pricing the same as Sonnet 4.5: $3 per million input tokens and $15 per million output tokens. It is the default model for Free and Pro users on claude.ai and Claude Cowork, and is also reachable through the API.

How large is the context window?

One million tokens, offered in beta. Anthropic says that is enough to place an entire codebase, long contracts or many research papers in a single request, and that reasoning quality holds up across that much context.

Is it better than Opus 4.5?

Anthropic reports that developers with early access often preferred it to Opus 4.5, and in Claude Code testing it was chosen over that model 59% of the time. The company still describes Opus 4.5 as its frontier model from November 2025.

What safety work has been done?

Anthropic says extensive evaluations found it as safe as, or safer than, its other recent Claude models, with strong safety behaviour and no major misalignment concerns. Prompt injection resistance improved over Sonnet 4.5 and is reported as similar to Opus 4.6.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Claude Sonnet 4.6 yet, so the facts on this page come from the sources listed above: anthropic.com/news/claude-sonnet-4-6. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.