Anthropic announced on March 13, 2026 that the 1 million token context window is now generally available for Claude Opus 4.6 and Sonnet 4.6, with a pricing structure that distinguishes it from competitors: a single flat rate across the entire window, no long-context premium. Opus 4.6 runs $5/$25 per million input/output tokens; Sonnet 4.6 is $3/$15. A 900,000-token request costs the same per-token rate as a 9,000-token one. The GA release also bundles full rate limits at every context length, a sixfold expansion in media limits to 600 images or PDF pages per request, and automatic availability across Claude Platform, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry without requiring a beta header.
For Claude Code users, the change is immediate and practical. Max, Team, and Enterprise plan subscribers using Opus 4.6 now default to 1M context automatically, directly cutting the frequency of compaction events — the truncation of conversation history that had been a persistent frustration for developers running long agent sessions. Jon Bell, CPO of a company using Claude agents, reported a 15% decrease in compaction events since adopting the expanded window. Software engineer Anton Biryukov noted that Claude Code sessions spanning Datadog, Braintree, databases, and source code can easily consume 100,000 or more tokens, and the full 1M window allows those workflows to stay intact without losing critical context mid-session.
Anthropic's flat pricing is a structural break from how its major competitors handle long-context requests. According to developer documentation, both Google's Gemini 2.5 Pro and OpenAI's GPT-5.4 apply tiered pricing multipliers — roughly 2x input and 1.5x output — once sessions cross context thresholds of 200,000 and 272,000 tokens respectively. Anthropic itself operated under a similar structure during the Sonnet 4 beta, where a 2x input multiplier kicked in above 200,000 tokens. By eliminating that premium entirely at GA, Anthropic positions Claude as potentially cost-competitive at enterprise-scale workloads that regularly push into the hundreds of thousands of tokens — even where its nominal base rates are higher than rivals. That math is straightforward; what's less clear is whether Google and OpenAI treat their multipliers as revenue lines worth defending or as competitive liabilities they'll quietly drop.
On the performance side, Anthropic reports Opus 4.6 scores 78.3% on MRCR v2 — a multi-document retrieval and comprehension benchmark — at the 1M context length, claiming the top position among frontier models at that length. Developer reaction on Hacker News has been broadly positive, with multiple commenters saying context compaction had pushed them toward GPT-5.4 and Codex and that they plan to switch back. That migration signal, early as it is, suggests Anthropic's bet on flat pricing as a developer retention tool is already finding its mark.