Anthropic increases Claude Code and API usage limits

Anthropic published a blog post titled "Higher usage limits for Claude and a compute deal with SpaceX" on May 6, 2026, that lists three immediate product changes: doubling Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans; removing the peak hours limit reduction on Claude Code for Pro and Max accounts; and materially raising API rate limits for `Claude Opus` models, with a rate table in the announcement showing substantially higher per-minute token throughput for multiple tiers. The blog post attributes the increases to a new compute partnership with SpaceX, stating access to the Colossus 1 facility will provide more than 300 megawatts of additional capacity and over 220,000 NVIDIA GPUs "within the month." 9to5Google reports specific uplifts, citing a 1500% increase to Tier 1 maximum input tokens per minute and a 900% increase to maximum output tokens per minute for that tier. Anthropic also said it will expand infrastructure presence in Asia and Europe for enterprise customers.

Per Anthropic's announcement, the changes apply across Claude surfaces where relevant, with the most visible user-facing change being the longer rolling five-hour window for Claude Code sessions on paid plans listed above. The company-provided API table for Claude Opus models shows higher rate ceilings by tier; 9to5Google summarizes some of the largest percentage jumps reported. Anthropic also reiterated that it trains and runs Claude using a mix of hardware, naming AWS Trainium, Google TPUs, and NVIDIA GPUs, and listed other compute arrangements alongside the SpaceX agreement.

Companies that secure large-scale, lower-cost compute often translate that capacity into higher per-user throughput and relaxed throttling, because more concurrent GPU resources reduce the need for aggressive rate-limiting. For practitioners, higher per-minute token limits typically reduce engineering workarounds such as aggressive request batching, complex client-side retry logic, or session-splitting to avoid hitting rolling-window caps. Increased API throughput also changes cost and latency trade-offs when moving from synchronous calls to longer, higher-token responses.

Industry reporting in March and April documented growing developer friction with Claude Code limits and promotional mitigations, including a March off-peak doubling promotion that Anthropic ran for two weeks, and user complaints covered by the BBC about hitting limits and unexpected token consumption. The SpaceX compute announcement is notable because access to unusually large, rapidly provisioned GPU capacity changes the supplier landscape for model hosting and could accelerate how vendors scale user-facing quotas. Observers should treat the compute deal and the limit increases as operational capacity news, not as a product roadmap signal from Anthropic beyond what the company published.

• •Adoption signals, such as changes in reported latency or fewer user complaints about being rate-limited.
• •Whether Anthropic publishes updated, detailed API rate tables and billing implications for enterprise customers.
• •How competitors and major cloud providers respond on pricing, quotas, and compute partnerships.

Sources for the reported facts include Anthropic's May 6, 2026 blog post and supporting documentation, 9to5Google's coverage summarizing percentage increases, the March support article on a prior off-peak promotion, and earlier BBC reporting on user issues with Claude Code.