Anthropic introduces Claude Sonnet 4.6 with 1M token context beta while holding API pricing flat

What Anthropic announced

Anthropic announced Claude Sonnet 4.6 on February 17, 2026 and described it as its most capable Sonnet model so far. The release frames Sonnet 4.6 as a broad capability upgrade rather than a narrow benchmark improvement, citing gains across coding, computer use, long-context reasoning, agent planning, knowledge work, and design workflows. A key technical headline is a 1M token context window in beta, which materially expands the type of document-heavy and multi-step tasks teams can run in a single session.

Rollout scope is also immediate. Anthropic says Sonnet 4.6 is now the default model for Free and Pro users in claude.ai and Claude Cowork. On the API side, Anthropic kept pricing unchanged versus Sonnet 4.5, starting at $3/$15 per million tokens. That combination of higher capability at stable price points is likely to matter for enterprises optimizing model mix, because it reduces pressure to escalate routine workloads to higher-cost tiers.

Performance and safety framing

In the release, Anthropic reports that early-access developers preferred Sonnet 4.6 to its predecessor by a wide margin, and in some practical tasks even preferred it to Claude Opus 4.5 from November 2025. The company highlights economically relevant office tasks and stronger computer-use behavior as areas where users see direct gains, suggesting Sonnet is being pushed further into production-grade agentic operations rather than remaining a mid-tier model for lighter prompts.

Anthropic also foregrounds risk management. The announcement explicitly discusses prompt injection risks in computer-use scenarios and points to mitigation work and safety evaluations. Its published safety conclusion is that Sonnet 4.6 appears as safe as, or safer than, recent Claude releases. For buyers in regulated sectors, that balance between expanded capability and explicit safeguards is as important as raw benchmark results.

Why it matters now

The strategic signal is clear: Anthropic is compressing the gap between mainstream and premium model utility. If Sonnet-class pricing can cover more complex coding and operational tasks, teams may redesign deployment economics, reserving top-tier models for edge cases only. The next validation checkpoint is real-world durability: latency under long context, consistency in multi-step tool use, and reliability under adversarial prompt conditions at scale.