Released February 17, 2026 · Anthropic
The Model That Changed The Math
Opus-level intelligence. Sonnet pricing. A full upgrade across coding, computer use, reasoning, agents, and design — at exactly zero extra cost.
FEB 17 2026 · API: claude-sonnet-4-6
The most capable mid-tier AI model ever released — and it's now your default.
Claude Sonnet 4.6 is the latest — and most capable — model in Anthropic's Sonnet line. Released February 17, 2026, it represents a full upgrade across six pillars: coding, computer use, long-context reasoning, agent planning, knowledge work, and design.
What makes it remarkable is not just that it's better at everything. It's that it delivers capability that previously required reaching for Anthropic's flagship Opus class — at a price that hasn't changed since Sonnet 4.5. Same $3 / $15 per million tokens. Materially smarter in every measurable dimension.
If you're on a Free or Pro plan at claude.ai or Claude Cowork, Sonnet 4.6 is already your default model. You've already upgraded — for free.
In blind head-to-head tests inside Claude Code, developers preferred Sonnet 4.6 over the previous flagship model (Opus 4.5) 59% of the time — despite Opus 4.5 being a tier above and more expensive. A mid-tier model beating the previous flagship is historically rare.
Capability Overview
Sonnet 4.5 vs. Sonnet 4.6 across all six pillars (illustrative, index 0–100)
Based on benchmark data and reported improvements. Computer use: OSWorld-Verified. Coding: SWE-bench Verified. Math: Box/enterprise evaluations.
79.6% on SWE-bench Verified. Preferred by developers 70% of the time over its own predecessor. Less overengineering, fewer hallucinations, better context retention in long sessions.
Explore →72.5% on OSWorld-Verified — nearly 5× higher than where the Sonnet line started just 16 months ago. Real software. No special APIs. 94% accuracy in insurance workflows.
Explore →1 million token context window (beta). Reasons — not just retrieves — across entire codebases, legal contracts, or dozens of research papers simultaneously.
Explore →Nearly tripled earnings vs Sonnet 4.5 in Vending-Bench Arena — a competitive, multi-model business simulation. Adaptive thinking decides when deep reasoning is needed.
Explore →89% math accuracy (up from 62%). Matches Opus 4.6 on OfficeQA — enterprise document comprehension. +15 percentage points on Box's heavy reasoning Q&A benchmark.
Explore →Independently described by multiple customers as having "perfect design taste." Better layouts, animations, and visual outputs — with fewer iteration rounds needed.
Explore →In October 2024, Anthropic called computer use "still experimental — at times cumbersome and error-prone." The score then was 14.9%. Sonnet 4.6 scores 72.5%. That's a 57.6 percentage point leap in sixteen months.
In the Vending-Bench Arena, where AI models compete to run a simulated business, Sonnet 4.6 developed its own novel strategy — invest heavily for 10 months, then pivot sharply to profit. It wasn't instructed to do this. It figured it out.
Math accuracy jumped from 62% to 89% — a 27 percentage point improvement. For context, that's the difference between a C-grade and an A in most academic grading curves.
A 1 million token context window can hold roughly 750,000 words — that's approximately 10 full-length novels, or the entire Linux kernel source code, in a single conversation.
Sonnet 4.6 is 40% cheaper per token than Opus 4.6 — yet in coding evaluations, developers preferred it over the previous flagship (Opus 4.5) more than half the time. The era of paying more for better code is over.
Anthropic's safety researchers described Sonnet 4.6 as having "a broadly warm, honest, prosocial, and at times funny character." That's not marketing copy — that's a formal safety evaluation finding.
Every benchmark with full charts, methodology notes, and historical context. For the data-driven.
⚡Deep dives into each of the six upgrade pillars with real-world examples and tutorial guidance.
🧬The full Sonnet family timeline — where each model came from and what it added to the line.
🗣️Seventeen executives from GitHub, Replit, Cursor, Zapier, Rakuten, and more share what they found.
🚀Where to access Sonnet 4.6, pricing details, API setup, and platform-specific guidance.
📄Read the full launch announcement from Anthropic with all official benchmark tables and quotes.