What Happened
Anthropic introduced 1M token context windows for Claude before OpenAI did the same. But as of March 2026, the rollout tells a different story. OpenAI's GPT-5.4 now offers 1M context to all users, while Claude still restricts it to certain accounts - even paying Max 20x subscribers don't always have access.
The gap is notable because Anthropic was the pioneer here. Claude's extended context was a genuine differentiator when it launched. Now the company that got there first is being outpaced on availability by the company that got there second.
Why It Matters
For anyone working with large codebases, long documents, or multi-file analysis, context window size directly affects what you can do in a single conversation. A 1M token window means you can feed in an entire repository, a full legal contract set, or months of data without chunking and losing coherence.
But capability only matters if you can actually use it. If you're paying for Claude's Max tier and still can't reliably access 1M context, you're paying premium prices for a feature that OpenAI now gives to its broader user base. That shifts the calculus for teams choosing between platforms.
This also affects developers building on these APIs. If you're designing a workflow around large context and need predictable availability, the platform that consistently delivers matters more than the one that technically supports it.
Our Take
This looks like an infrastructure scaling problem, not a strategic choice. Running 1M context inference is expensive. Each request uses significantly more compute than a standard conversation, and Anthropic is a smaller company than OpenAI with fewer resources to throw at GPU capacity. They likely can't afford to let every user slam their infrastructure with million-token requests simultaneously.
OpenAI has the compute advantage here. Microsoft's backing gives them access to massive Azure clusters that Anthropic simply can't match yet. So while Anthropic had the better model capability earlier, OpenAI had the infrastructure to democratize it faster.
The real question is whether this matters for most users. Very few workflows actually need 1M tokens in practice. Most people hitting context limits are doing so around 100-200K tokens. The 1M ceiling is impressive on paper, but it's more of a benchmark flex than a daily-use feature for the majority of users.
That said, perception matters. When users see "1M context" on OpenAI's feature list and "limited availability" on Claude's, it shapes purchasing decisions regardless of actual usage patterns. Anthropic needs to either roll this out broadly or stop leading with it as a differentiator. Half-available features create more frustration than no feature at all.
If extended context is critical to your workflow right now, GPT-5.4 is the more reliable bet for consistent access. If you prefer Claude's reasoning quality and can work within its standard context limits, nothing has changed. Pick based on what you actually use, not what the spec sheet promises.