Anthropic Releases Claude Sonnet 4.6: The Most Capable Sonnet Model Yet

February 17, 2026 — Anthropic has launched Claude Sonnet 4.6. This represents a major upgrade to its AI model lineup. The improvements span coding, computer use, long-context reasoning, and design capabilities.

Furthermore, the new model is now live for all users. It’s the default option in claude.ai and Claude Cowork for Free and Pro plans. Additionally, API pricing remains unchanged from Sonnet 4.5 at $3 per million input tokens and $15 per million output tokens.

Major Performance Upgrades Across the Board

Claude Sonnet 4.6 brings what Anthropic calls a “full upgrade” of capabilities. Early access developers overwhelmingly prefer the new version. In fact, internal testing shows users chose Sonnet 4.6 over Sonnet 4.5 approximately 70% of the time.

Moreover, users preferred Sonnet 4.6 over Claude Opus 4.5 59% of the time. That’s significant because Opus 4.5 was Anthropic’s flagship model from November 2025.

According to Anthropic, tasks that previously required Opus-class models can now run on Sonnet 4.6. This includes real-world office tasks. Consequently, users get high-end performance at a lower price point.

The model includes a 1 million token context window in beta. This is enough space for entire codebases, lengthy contracts, or dozens of research papers. More importantly, Sonnet 4.6 can reason effectively across all that context. As a result, it enables much better long-term planning.

Computer Use: From Experimental to Production-Ready

Anthropic first introduced computer use in October 2024. At the time, the company called it “experimental” and “error-prone.” However, sixteen months later, the progress is remarkable.

OSWorld is the standard test for AI computer use. It presents hundreds of tasks across real software like Chrome, LibreOffice, and VS Code. On this benchmark, Sonnet models have shown steady gains.

Early Sonnet 4.6 users now report human-level ability in complex tasks. These include navigating spreadsheets, filling out multi-step web forms, and working across multiple browser tabs.

Admittedly, the model still lags behind skilled human computer users. Nevertheless, the rate of progress suggests more capable models are coming soon.

Security is also improving. Specifically, Sonnet 4.6 better resists prompt injection attacks. These attacks happen when malicious actors hide instructions on websites. In safety tests, Sonnet 4.6 showed major improvements over Sonnet 4.5. In addition, it performs similarly to Opus 4.6 in blocking these threats.

Coding: Less Overengineering, Better Follow-Through

Developers testing Claude Code noticed several improvements. First, Sonnet 4.6 is less prone to overengineering. Second, it shows better instruction following compared to previous models.

Users noted specific strengths:

• The model reads context before modifying code
• It combines shared logic rather than duplicating it
• It produces fewer false success claims
• It generates fewer errors
• It shows more consistent follow-through on multi-step tasks

Additionally, enterprise customers highlighted strong improvements in frontend code and financial analysis. Visual outputs are “notably more polished.” They feature better layouts, animations, and design sense than previous models.

What Industry Leaders Are Saying

The release has generated strong praise from major technology companies. Here’s what leaders are saying:

Michele Catasta, President of Replit: “The performance-to-cost ratio of Claude Sonnet 4.6 is extraordinary. It’s hard to overstate how fast Claude models have been evolving.”

Michael Truell, Co-founder and CEO of Cursor: “Claude Sonnet 4.6 is a notable improvement over Sonnet 4.5. This includes long-horizon tasks and more difficult problems.”

Joe Binder, VP of Product at GitHub: “Out of the gate, Claude Sonnet 4.6 excels at complex code fixes. This is especially true when searching large codebases.”

Hanlin Tang, CTO of Neural Networks at Databricks: “Claude Sonnet 4.6 matches Opus 4.6 performance on OfficeQA. It’s a meaningful upgrade for document comprehension workloads.”

Safety and Character Assessment

Anthropic conducted extensive safety tests of Sonnet 4.6. The results show the model is “as safe as, or safer than” other recent Claude models.

Safety researchers concluded that Sonnet 4.6 has a “broadly warm, honest, prosocial, and at times funny character.” Furthermore, it shows “very strong safety behaviors.” Importantly, there are “no signs of major concerns around high-stakes forms of misalignment.”

New Features and Product Updates

Beyond the core model improvements, Anthropic announced several product updates.

Claude Developer Platform enhancements:

• Adaptive thinking and extended thinking support
• Context compaction (beta) that summarizes older context automatically
• Web search and fetch tools now filter and process search results with code
• Code execution, memory, and programmatic tool calling are now available to everyone

Claude in Excel updates:

• MCP connector support for external data providers
• Works with S&P Global, LSEG, Daloopa, PitchBook, Moody’s, and FactSet
• Available on Pro, Max, Team, and Enterprise plans

Free tier upgrades:

• Sonnet 4.6 is now the default model
• Free tier now includes file creation, connectors, skills, and compaction

Availability and Pricing

Claude Sonnet 4.6 is available immediately across multiple platforms:

• All Claude plans (Free, Pro, Max, Team, Enterprise)
• Claude Cowork
• Claude Code
• The Claude API (model string: claude-sonnet-4-6)
• Major cloud platforms including Amazon Bedrock and Google Cloud’s Vertex AI

API pricing remains the same as Sonnet 4.5:

• $3 per million input tokens
• $15 per million output tokens

For the deepest reasoning tasks, Anthropic still recommends Opus 4.6. This includes codebase refactoring and coordinating multiple agents. It’s also best for problems where precision is critical.

What This Means for AI Development

The release of Sonnet 4.6 is a significant milestone. It makes advanced AI capabilities more accessible. By delivering near-Opus performance at Sonnet pricing, Anthropic has lowered the cost barrier.

The improvements in computer use are particularly striking. In just sixteen months, these capabilities moved from experimental to production-ready. This suggests AI agents that can navigate standard software are becoming practical reality.

For developers and businesses using Claude, the recommendation is clear: test Sonnet 4.6 in your workflows. Many use cases that previously required Opus may now work with Sonnet 4.6. Moreover, you’ll get this performance at a fraction of the cost.

About Anthropic

Anthropic is an AI safety company that builds frontier AI systems. This includes the Claude family of models. The company focuses on developing AI that is helpful, harmless, and honest. Additionally, they emphasize safety research and responsible deployment practices.

For more information and full technical documentation, visit anthropic.com/news/claude-sonnet-4-6. Alternatively, check the Claude Developer Platform documentation.