Anthropic tightens Claude usage limits

Anthropic’s latest Claude limits story is not a simple tightening or loosening. On May 6, the company said it had raised usage limits for Claude Code and the Claude API, including doubling Claude Code’s five-hour rate limits for Pro, Max, Team and seat-based Enterprise plans, while also removing peak-hours limit reductions for Pro and Max accounts. (anthropic.com) Anthropic tied those changes to a new compute partnership with SpaceX that it said would add more than 300 megawatts of capacity and over 220,000 Nvidia GPUs within a month. (ramp.com) (anthropic.com) April 23 is the other key date. Anthropic said that, after complaints about Claude Code quality, it had reset usage limits for all subscribers while reversing or fixing several product changes that had hurt performance in Claude Code, the Claude Agent SDK and Claude Cowork. The company said the API was not affected. Those two updates sit alongside a broader market fight over coding agents. OpenAI said in February that, for a limited time, Codex would be included with ChatGPT Free and Go, while paid users would get doubled Codex rate limits. (anthropic.com) OpenAI’s current pricing page says some Codex usage is still governed by separate limits that may adjust based on demand. ### If Anthropic raised limits on May 6, why are people talking about tighter access? April 23 explains much of the confusion. Anthropic said it had traced quality complaints to three changes made between March 4 and April 16, including lowering Claude Code’s default reasoning effort, a bug that repeatedly cleared older session thinking, and a prompt change that hurt coding quality. (anthropic.com) As part of that response, Anthropic said it was resetting usage limits for all subscribers. Ramp economist Ara Kharazian said on May 13 that recent weeks had brought “frequent outages, rate limits, and increasing dissatisfaction with results” for some Claude users, while also noting that Anthropic had responded quickly and that the SpaceX deal should ease near-term compute constraints. (openai.com) That was Kharazian’s interpretation of Ramp’s business-spend data, not a company disclosure from Anthropic. ### What did Anthropic actually change for Claude Code and the API? Anthropic’s May 6 post lists three concrete changes. First, Claude Code’s five-hour rate limits were doubled for Pro, Max, Team and seat-based Enterprise users. (anthropic.com) Second, peak-hours limit reductions were removed for Pro and Max accounts. Third, API rate limits for Claude Opus models were raised. More than 300 megawatts of new capacity is the number Anthropic attached to the SpaceX agreement. The company said that capacity would “directly improve capacity for Claude Pro and Claude Max subscribers,” and added that it also has compute agreements with Amazon, Google, Broadcom, Microsoft, Nvidia and Fluidstack. (ramp.com) ### What do the Ramp numbers say about business adoption? Ramp said on May 13 that Anthropic passed OpenAI in business adoption for the first time in its AI Index. Anthropic’s adoption rose 3.8 percentage points in April to 34.4% of businesses on Ramp, while OpenAI fell 2.9 points to 32.3%, according to the report. (anthropic.com) Overall AI adoption rose to 50.6%. Ara Kharazian also warned that the lead should not be read as settled. He wrote that Anthropic faces headwinds from cost, compute constraints and user dissatisfaction, and said some fast-growing vendors on Ramp’s platform are inference providers offering cheaper open-source models. (anthropic.com) ### Where does Mythos fit into this story? The U.K. AI Security Institute said on April 13 that Anthropic’s Claude Mythos Preview showed “continued improvement” in capture-the-flag tasks and “significant improvement” on multi-step cyber-attack simulations. (ramp.com) AISI said Mythos Preview succeeded 73% of the time on expert-level tasks in its cyber CTF suite. AISI also said Mythos Preview could execute multi-stage attacks on vulnerable networks and autonomously discover and exploit vulnerabilities in controlled evaluations. (ramp.com) Those findings speak to model capability, but they also underscore why compute, access controls and pricing are becoming operational questions for model providers shipping agentic products. That last point is an inference from Anthropic’s and OpenAI’s product changes, supported by the usage-limit and pricing updates each company has published. (aisi.gov.uk) ### What should readers watch next? May 14 is the latest date on Anthropic’s newsroom page tied to enterprise rollout. The company said PwC is deploying Claude for client and internal work, and Anthropic also listed “Introducing Claude for Small Business” on May 6 and “Agents for financial services” on May 5 among recent product announcements. May 31 is one date on OpenAI’s side. OpenAI’s Codex pricing page says the company is doubling normal Codex usage on the $100-per-month Pro tier until May 31, 2026, while Anthropic says the SpaceX capacity should come online within a month of its May 6 announcement. (aisi.gov.uk) (developers.openai.com) (anthropic.com)