Anthropic raises Claude API rate limits and updates how tiers work
Anthropic says it has raised Claude Platform API rate limits for all users and simplified usage tiers. The company’s developer announcement says tiers are no longer based on API spend, and points users to updated Claude Platform rate limit documentation for current per-model RPM and token throughput limits.
Anthropic has updated Claude Platform API rate limits and how its usage tiers work. In a developer update, Anthropic says it has raised Claude Platform API rate limits for all users and simplified the tiers, with tiers "no longer based on API spend." (ClaudeDevs on X)
The company also says the latest Sonnet and Haiku models now provide 5× higher rate limits at the highest tier, and it points developers to the official Claude Platform rate limit documentation for the current tables and mechanics. (ClaudeDevs on X, Claude Platform docs)
What changed
1) Tier mechanics: what we can verify
- Anthropic states that it has "simplified the tiers" and that they are "no longer based on API spend." (ClaudeDevs on X)
- The Claude Platform documentation describes tiers as automatic and usage-based: "Your organization is placed on a tier automatically and can move to a higher tier over time as you use the API." (Claude Platform docs)
Important note: The documentation page still includes a "Spend limits" section with monthly spend caps for several tiers, so we are not inferring additional details beyond Anthropic’s explicit statement in the X post. (Claude Platform docs)
2) Higher rate limits on the newest Sonnet + Haiku
Anthropic’s developer announcement says: "The latest Sonnet and Haiku models now provide 5x higher rate limits at the highest tier." (ClaudeDevs on X)
Anthropic does not include the before/after numbers in the post itself, so treat the official rate limit tables as the source of truth for current limits and implementation details. (Claude Platform docs)
Practical takeaways for developers
- Re-check your capacity assumptions. If you have been throttled by RPM or token throughput on Sonnet or Haiku, this change may materially increase headroom at the top tier — but you should verify your current tier and limits in the Claude Console. (ClaudeDevs on X, Claude Platform docs)
- Design for 429s anyway. Anthropic’s docs emphasize that if you exceed any limit you receive a 429 plus a
retry-afterheader, so robust backoff remains required even with higher limits. (Claude Platform docs) - Know which limits you’re hitting. The docs describe rate limits in RPM plus input/output tokens per minute, and they document response headers like
anthropic-ratelimit-requests-remainingandanthropic-ratelimit-output-tokens-remainingto help you observe throttling in production. (Claude Platform docs)
Sources
- ClaudeDevs announcement (July 2, 2026): https://x.com/ClaudeDevs/status/2072818299361263778
- Claude Platform docs, “Rate limits”: https://platform.claude.com/docs/en/api/rate-limits
Frequently Asked Questions
What did Anthropic change about Claude API rate limits?
Anthropic says it raised Claude Platform API rate limits for all users. It also says the newest Sonnet and Haiku models provide 5× higher rate limits at the highest tier.
How do Claude Platform usage tiers work now?
Anthropic says tiers are no longer based on API spend and that it simplified the tiers. The Claude Platform docs describe tiers as automatic and usage-based, with organizations moving to higher tiers over time as they use the API.
Where can developers verify their exact RPM and token limits?
Anthropic points developers to the Claude Platform rate limit documentation and the Claude Console Limits page, which show per-model RPM plus input/output tokens per minute limits and provide a way to request higher limits.