Anthropic has released Claude Sonnet 4.6, an AI model that is shaking up the industry by delivering premium-tier performance at a fraction of the cost. According to internal testing, users preferred Sonnet 4.6 over its predecessor 70% of the time, and even over the flagship Opus 4.5 model 59% of the time.
What is Claude Sonnet 4.6 and why it matters
Claude Sonnet 4.6 is Anthropic's new AI model, arriving as a direct upgrade to Sonnet 4.5. What makes it revolutionary is straightforward: it delivers Opus-comparable performance at just $3 per million input tokens and $15 per million output tokens, roughly one-fifth of Opus pricing.
This means developers and businesses that previously needed the most expensive model for complex tasks can now achieve similar results without breaking their budget.
Key improvements over Sonnet 4.5
The improvements are substantial:
- 1M token context window (in beta): double any previous Sonnet, allowing processing of massive documents in a single pass
- Superior coding: better consistency, less overengineering, and improved instruction-following
- Long-context reasoning: 15 percentage points better than Sonnet 4.5 on complex reasoning Q&A
- Improved computer use: 72.5% on OSWorld-Verified, up from 14.9% when the capability launched in 2024
- Fewer hallucinations: significant reduction in fabricated responses and false success claims
- Better visual design: more polished outputs with improved design sensibility
How it compares to OpenAI's GPT-5.2
Anthropic claims Sonnet 4.6 outperforms OpenAI's GPT-5.2 on knowledge work benchmarks, particularly in finance and legal fields. In an insurance company benchmark, the model achieved 94% accuracy.
This is particularly relevant given that OpenAI just retired GPT-4o and other legacy models, consolidating GPT-5.2 as their standard. The competition between both companies has never been tighter.
Availability and pricing
Sonnet 4.6 is already available as the default model for Free and Pro users on claude.ai and Claude Cowork. It is also available through the API and major cloud platforms (AWS Bedrock, Google Cloud Vertex AI).
- Input: $3 per million tokens
- Output: $15 per million tokens
What this means for developers
If you are a developer using AI for coding, this update is significant. Claude Code showed users prefer Sonnet 4.6 over the previous version in 7 out of 10 cases. The model generates cleaner code, follows instructions better, and makes fewer mistakes.
The 1M token window is also a game-changer for those working with large codebases: you can feed the model entire repositories and get contextualized responses.
Safety and reliability
Anthropic reports that Sonnet 4.6 has "a broadly warm, honest, prosocial character, very strong safety behaviors, and no signs of major concerns." Safety evaluations show significant improvements in prompt injection resistance compared to Sonnet 4.5.
The AI landscape in February 2026
This release comes at a time when AI competition is at its most intense. OpenAI consolidated GPT-5.2, Google is preparing announcements for I/O 2026, and now Anthropic proves that Opus-level performance can be accessible to everyone. The question is no longer whether AI is useful for programming, but how much you are overpaying by not using the right model.
