Anthropic's Sonnet 4.6 Closes the Gap: Mid-Tier AI Now Matches Last Year's Flagship
News Summary
Anthropic officially launched Claude Sonnet 4.6 on Tuesday, February 17, 2026 (10:00 AM PST), marking the company's second major AI model release in under two weeks. The rollout follows the debut of Claude Opus 4.6 on February 5, 2026 (13 days prior), underscoring the breakneck development pace in the frontier AI industry.
The new model becomes the default for both Free and Pro plan users on Claude.ai, Claude's web interface, and the newly released Claude Cowork desktop app — at no additional cost.
What's New: Key Upgrades
1. Expanded Context Window — 1 Million Tokens (Beta)
The headline feature is a 1 million token context window, now available in beta, doubling the largest context previously offered in Sonnet-class models. Anthropic describes this as sufficient to hold "entire codebases, lengthy contracts, or dozens of research papers in a single request" — a capability that had previously been exclusive to Opus 4.6.
2. Dramatically Improved Coding Capabilities
Anthropic says developers with early access "prefer Sonnet 4.6 to its predecessor by a wide margin" — and in many cases even prefer it to the previous flagship Claude Opus 4.5 (November 2025). Improvements include stronger consistency, better instruction-following, long-context code reasoning, and smarter debugging logic. Claude Code, Anthropic's agentic coding CLI, preferred Sonnet 4.6 outputs roughly 70% of the time over prior models.
3. Advanced Computer Use Skills
On the OSWorld-Verified benchmark — which tests how well AI agents navigate real desktop and web software — Sonnet 4.6 scored 72.5%, up from 61.4% on the previous Sonnet model, and even surpassing Claude Opus 4.5's 66.3%. It now trails Anthropic's own flagship Opus 4.6 by only 0.2 percentage points on this metric.
4. Agentic Office Task Performance
Anthropic positions Sonnet 4.6 as capable of completing complex office tasks that previously demanded Opus-class models — including navigating spreadsheets, filling multi-step web forms, and managing multi-tab browser research sessions. Box CTO Ben Kus confirmed Sonnet 4.6 outperformed its predecessor on deep reasoning Q&A tasks by 15 percentage points in enterprise document workflows.
5. Stronger Safety Profile
According to Anthropic's internal safety evaluations, Sonnet 4.6 shows major improvements in resistance to hallucinations and external manipulation compared to Sonnet 4.5, and performs similarly to Opus 4.6 in safety benchmarks. Anthropic describes the model as having "a broadly warm, honest, prosocial, and at times funny character."
Pricing: Same as Before
Despite the significant capability upgrades, Anthropic held pricing flat at $3 per million input tokens and $15 per million output tokens via the API — identical to Sonnet 4.5. Claude Pro for individuals remains $20/month (or $17/month with annual billing). Claude Opus 4.6 continues to be priced at $5/$25 per million tokens.
Competitive Context
The release intensifies competition with OpenAI and Google. Internal benchmarks show Sonnet 4.6 outperforming Google Gemini 3 Pro and OpenAI GPT 5.2 in agentic financial analysis and office workloads. In some categories, Sonnet 4.6 even surpassed Anthropic's own Opus 4.6 — a notable internal leapfrog.
Anthropic's enterprise momentum is accelerating: customers spending over $100,000 annually on Claude have grown 7x year-over-year, and those spending $1 million+ annually grew from approximately 12 two years ago to more than 500 today. The company was recently valued at $380 billion following a $30 billion funding round.
What Comes Next
An updated Haiku model (Anthropic's smallest and fastest tier) is expected in the coming weeks, likely completing the full 4.6 model family refresh. A Windows version of Claude Cowork is also in development, currently available only on macOS.
Bottom Line
Claude Sonnet 4.6 represents a strategic move to make Opus-class performance accessible at mid-tier prices — democratizing advanced AI for everyday enterprise use. As the AI model race shifts from raw benchmark chasing toward operational efficiency, Anthropic's bet on practical throughput, safety, and cost optimization may well define the next phase of enterprise AI adoption.