Price Unchanged, Performance Soars: Anthropic’s Sonnet 4.6 Offers Free 1M‑Token Context

Core Upgrade Overview

Anthropic announced Sonnet 4.6 on Feb 17, doubling the context window to a beta 1 million‑token limit that is free for all users while keeping the input price at $3 per million tokens.

1 Million‑Token Context: Top‑Tier Capability for Everyone

One million tokens correspond to roughly 750 k Chinese characters—about the length of a full novel. This allows an entire codebase, a year’s meeting minutes, or dozens of papers to be fed to the model in a single request, enabling precise retrieval and full‑text understanding without splitting.

The capability was previously exclusive to the flagship Opus model or behind a high‑price paywall; Sonnet 4.6 makes it available in beta to API developers and Claude Code users, while the standard Claude AI web UI remains limited.

Half‑Price Flagship‑Level Performance

Pricing remains unchanged: $3 / million input tokens and $15 / million output tokens. By contrast, Opus 4.6 costs $5 / million input and $25 / million output, meaning Sonnet 4.6 costs only 60 % of the flagship.

Performance improvements are reported across four dimensions:

Programming (SWE‑Bench) : higher code‑generation, bug‑fix, and full‑stack refactoring scores with more actionable solutions and fewer hallucinations.

Computer Operations (OS World) : cross‑software, multi‑step automation such as batch spreadsheet processing, file organization, and end‑to‑end approval workflows.

Complex Task Planning (ARC‑AGI‑2) : achieved a 60.4 % high‑score, delivering qualitatively better planning and closed‑loop execution for AI agents and automated workflows.

Real‑World Case : a development team switched its code‑review AI agent from Opus 4.5 to Sonnet 4.6, reducing API costs by 80 % while maintaining comparable code‑fix success rates, dramatically improving cost‑effectiveness for small‑to‑mid‑size teams.

Who Should Switch Now?

Individual users : zero‑cost migration on Claude AI; large documents such as dozens of pages of contracts or papers can be processed directly.

Developers : building AI applications, agents, or workflows with Sonnet 4.6 offers flagship‑level performance at 60 % of Opus cost, making it the preferred choice for solo developers and small teams.

Enterprise users : the model meets performance, cost, and safety requirements for contract review, financial analysis, internal knowledge bases, and office automation, delivering near‑flagship efficiency without the flagship price.

Quick Highlights

Claude’s website defaults to Sonnet 4.6—just open and use.

API model identifier: claude-sonnet-4-6, pricing unchanged.

Beta 1 million‑token context is open to API developers and Claude Code users.

In early 2026 Anthropic released both Opus 4.6 and Sonnet 4.6, systematically lowering the barrier to top‑tier LLM capabilities.