Overview
Claude Sonnet 4.6 is Anthropic's most capable Sonnet model to date, released on February 17, 2026. It closes the gap with Opus-tier performance across coding, computer use, and long-context reasoning — while keeping the same $3/$15 per million token pricing as Sonnet 4.5. For developers who previously needed to reach for an Opus-class model, Sonnet 4.6 now handles those workloads at a more practical price point. It also introduces a 1M token context window in beta and becomes the new default model on Claude.ai Free and Pro plans.
What's New
Coding: Closing the Gap with Opus
Claude Sonnet 4.6 delivers a significant jump in coding quality. In Claude Code, internal testing showed users preferred Sonnet 4.6 over Sonnet 4.5 roughly 70% of the time — and even preferred it over Opus 4.5 59% of the time. Developers report that Sonnet 4.6 reads context more effectively before modifying code, consolidates shared logic rather than duplicating it, and produces fewer false claims of success or hallucinations across long sessions.
On SWE-bench Verified, Sonnet 4.6 achieves 80.2% with a prompt modification (10-trial average per Anthropic's methodology), placing it among the top frontier models for real-world software engineering tasks. For Cursor, Replit, GitHub, and other agentic coding platforms, early evaluations confirm Sonnet 4.6 outperforms its predecessor on orchestration tasks, complex bug-fixing, and large codebase navigation.
Computer Use: Human-Level on Everyday Tasks
Anthropic first introduced computer use in October 2024. Sonnet 4.6 represents a major step forward on OSWorld, the standard benchmark for AI computer use that tests real software — Chrome, LibreOffice, VS Code, and more — without special APIs.
Early users report human-level capability on tasks like navigating complex spreadsheets or completing multi-step web forms across multiple browser tabs. Sonnet 4.6 also demonstrates major improvements in resisting prompt injection attacks (where malicious actors hide instructions in websites), performing similarly to Claude Opus 4.6 in safety evaluations.
1M Token Context Window (Beta)
Sonnet 4.6 introduces a 1 million token context window in beta — enough to hold entire codebases, lengthy legal contracts, or dozens of research papers in a single request. More importantly, Sonnet 4.6 reasons effectively across the full context, not just retrieving information but enabling long-horizon planning. This was demonstrated on Vending-Bench Arena, where Sonnet 4.6 developed a novel strategy: investing in capacity for the first ten simulated months, then pivoting sharply to profitability — and finishing well ahead of competing models.
Instruction Following & Reliability
Anthropic reports early users and partners found that Sonnet 4.6 requires fewer revision rounds to reach production quality. It is significantly less prone to overengineering and "laziness," and meaningfully better at following multi-step instructions without drift. Frontend outputs are notably more polished — multiple early access customers independently described visual outputs as having better layouts and design sensibility than previous models.
Extended Thinking & Context Compaction
On the Claude Developer Platform, Sonnet 4.6 supports both adaptive thinking and extended thinking. It also supports context compaction in beta, which automatically summarizes older context as conversations approach limits — increasing effective context length for long-running agentic tasks.
Pricing & Plans
Claude Sonnet 4.6 pricing remains unchanged from Sonnet 4.5:
| Plan | Price |
|---|---|
| API – Input | $3 per million tokens |
| API – Output | $15 per million tokens |
| Free (Claude.ai) | Available — Sonnet 4.6 is now the default model |
| Pro (Claude.ai) | Included — higher usage limits |
| Team / Enterprise | Contact Anthropic for volume pricing |
The free tier has been upgraded: it now includes file creation, connectors, skills, and compaction alongside Sonnet 4.6. For Claude in Excel users, MCP connectors are available on Pro, Max, Team, and Enterprise plans.
Note on 1M context pricing: The 1M token context window is in beta on the Claude Developer Platform only, and access depends on your organization's usage tier and custom limits. Input exceeding 200K tokens incurs long-context premium rates (approximately $6 input / $22.5 output per million tokens) instead of the standard $3/$15 rates.
Best For
- Developers using Claude Code or API integrations who need near Opus-level code quality without Opus pricing
- Teams automating enterprise software workflows with computer use (form filling, spreadsheet navigation, multi-tab browser tasks)
- Researchers and analysts processing large document sets — contracts, papers, codebases — that require reasoning across the full context
- Organizations building AI chatbot or agent products that previously required Opus for reliability
- Frontend engineers who need polished UI/UX output with fewer revision cycles
- Data and finance teams running complex multi-step reasoning tasks like document Q&A, contract routing, and CRM coordination
FAQ
How does Claude Sonnet 4.6 compare to Sonnet 4.5?
Sonnet 4.6 is a full upgrade over 4.5 across coding, computer use, long-context reasoning, agent planning, and design. In Claude Code, users preferred Sonnet 4.6 over Sonnet 4.5 roughly 70% of the time. Computer use on OSWorld shows consistent improvements. Pricing is unchanged at $3/$15 per million tokens.
Is the 1M token context window available to all users?
The 1M token context window is available in beta on the Claude Developer Platform (API) only, and requires your organization to have the appropriate usage tier or custom limits enabled. Input exceeding 200K tokens incurs long-context premium pricing (~$6/$22.5 per million tokens) rather than the standard $3/$15 rates. Anthropic recommends evaluating it for your use case before relying on it in production.
Can I use Claude Sonnet 4.6 for computer use tasks?
Yes. Sonnet 4.6 supports computer use and shows human-level capability on everyday tasks like spreadsheet navigation and multi-step web form completion. It also has significantly improved prompt injection resistance compared to Sonnet 4.5. See Anthropic's API docs for guidance on safe computer use deployments.
How does Sonnet 4.6 compare to AI code generator alternatives like GPT-5.3-Codex?
Both are strong options for agentic coding. Sonnet 4.6 achieves 80.2% on SWE-bench Verified (with prompt modification, per Anthropic) and is preferred by Claude Code users for consistency and instruction following. GPT-5.3-Codex targets computer-use tasks beyond coding. The best choice depends on your workflow and existing integrations.
When should I use Opus 4.6 instead of Sonnet 4.6?
Anthropic recommends Claude Opus 4.6 for tasks that demand the deepest reasoning — codebase refactoring, coordinating multiple agents in a workflow, and situations where getting it exactly right is paramount. For most coding, computer use, and long-context tasks, Sonnet 4.6 now approaches Opus-level performance at a lower cost.
Is Claude Sonnet 4.6 safe for enterprise use?
Anthropic ran extensive safety evaluations before release. Sonnet 4.6 has "a broadly warm, honest, prosocial, and at times funny character, very strong safety behaviors, and no signs of major concerns around high-stakes forms of misalignment." It also shows major improvement in prompt injection resistance for computer use. See Anthropic's system card for full safety evaluation details.
