AI Models Platforms

Anthropic Launches Claude Sonnet 4.6 With Near-Opus Performance

Anthropic's mid-tier model now matches flagship-class performance at $3/$15 per million tokens.

Andrés Martínez
Andrés MartínezAI Content Writer
February 18, 20262 min read
Share:
Abstract visualization of layered neural network architecture with glowing connections representing Claude Sonnet 4.6 model capabilities

Anthropic released Claude Sonnet 4.6 on Tuesday, a mid-tier model that its own early testers often preferred over the company's flagship Opus 4.5. Pricing stays at $3/$15 per million tokens, one-fifth of what Opus costs. The model is now the default for free and Pro users on claude.ai.

The coding gains are the headline here. In Claude Code testing, developers preferred Sonnet 4.6 over its predecessor 70% of the time, per Anthropic's own numbers. They even picked it over Opus 4.5 in 59% of head-to-head comparisons, citing fewer hallucinations and less overengineering. Those are company-reported figures, worth noting, but early partner reactions from Cursor, GitHub, and Replit back up the general direction.

Computer use saw the most dramatic jump. On the OSWorld benchmark, which tests AI navigating real software like Chrome and VS Code, Anthropic's Sonnet models climbed from 14.9% when the capability launched in October 2024 to 72.5% now. The model also picked up prompt injection resistance on par with Opus 4.6, a real concern for any agent clicking around the web unsupervised.

A 1M token context window (in beta) rounds out the release. Anthropic says the model reasons across that full context, not just stores it. On the Vending-Bench Arena, a simulated business competition, Sonnet 4.6 outperformed competitors by investing heavily early, then pivoting to profitability late.

Sonnet 4.6 is available now via the API as claude-sonnet-4-6. Opus 4.6 launched just 12 days ago, making this Anthropic's second major model drop in under two weeks.


Bottom Line

Anthropic's $3/$15 mid-tier model now matches or beats its own $15/$75 flagship from November on most coding and office tasks, according to company-reported testing.

Quick Facts

  • Pricing: $3 input / $15 output per million tokens (unchanged from Sonnet 4.5)
  • Context window: 1M tokens (beta)
  • OSWorld-Verified score: 72.5%, up from 14.9% at computer use launch (Oct 2024)
  • Developer preference: 70% over Sonnet 4.5, 59% over Opus 4.5 (company-reported)
  • API identifier: claude-sonnet-4-6
Tags:AnthropicClaude Sonnet 4.6AI modelscomputer usecoding AIlarge language modelsAI agents
Andrés Martínez

Andrés Martínez

AI Content Writer

Andrés reports on the AI stories that matter right now. No hype, just clear, daily coverage of the tools, trends, and developments changing industries in real time. He makes the complex feel routine.

Related Articles

Stay Ahead of the AI Curve

Get the latest AI news, reviews, and deals delivered straight to your inbox. Join 100,000+ AI enthusiasts.

By subscribing, you agree to our Privacy Policy. Unsubscribe anytime.

Claude Sonnet 4.6: Anthropic's Mid-Tier Model Matches Opus | aiHola