Check your Claude API tier: Anthropic raised Sonnet and Haiku rate limits

Anthropic raised rate limits across the Claude API and simplified its usage tiers. Confidence level: confirmed. The practical change is that Claude Sonnet and Claude Haiku rate limits now match Claude Opus at every usage tier, while tiers have been consolidated into Start, Build, and Scale.

What changed

Anthropic's June 26, 2026 Claude Platform release notes say Sonnet and Haiku now share the same rate-limit levels as Opus across every usage tier. The company also consolidated usage tiers into three names: Start, Build, and Scale.

Anthropic says most organizations move to a higher tier, no organization receives lower limits than before, and no action is required. The exact numbers still depend on the organization's current Claude Console limits, spend cap, model, and endpoint behavior.

Why this is early

The primary update is in Anthropic's own Claude Platform release notes, and Releasebot captured the same June 26 entry in its Claude Developer Platform feed. Anthropic's rate-limit documentation provides the stable reference for tier concepts and spend caps.

This is not a new model release. It matters because rate limits decide whether agents, batch jobs, coding tools, and customer-facing AI features fail cleanly or stall during peak usage.

Key takeaways

• Sonnet and Haiku now match Opus rate limits at each Claude API usage tier.
• Usage tiers are now Start, Build, and Scale.
• Anthropic says no action is required and no organization receives lower limits.
• Teams should still inspect their Claude Console because exact limits are account-specific.
• Agent workloads should retest retry, backoff, and queue behavior after the tier change.

Availability and access

The rate-limit change is live according to Anthropic's release notes. Developers can view their current tier and limits in the Claude Console, while the public docs explain the tier structure and spend-cap model.

Pricing is not the same thing as rate limits. Matching Sonnet and Haiku rate limits to Opus does not mean the models cost the same, behave the same, or fit the same latency profile. It only changes the ceiling available to organizations at a given tier.

Practical LinkLoot angle

If you run Claude-powered agents, this is a good moment to rerun throughput tests. Check whether jobs that previously hit Sonnet or Haiku limits now complete without throttling, then keep the old retry logic anyway.

For workflow builders, the useful upgrade is operational headroom. Higher limits can make multi-step agents less brittle, but they can also expose weak spend controls. Pair the change with budget alerts, queue caps, and per-task token ceilings. LinkLoot's AI agent tools guide is a good place to audit the rest of the stack.

What to verify before you act

• Open the Claude Console and record your current tier, RPM, ITPM, OTPM, and spend cap.
• Recheck model pricing before moving traffic between Opus, Sonnet, and Haiku.
• Load-test the exact agent path that used to hit 429s.
• Confirm retry and backoff behavior for streaming, batch, and long-running tool calls.
• Update internal runbooks so teams know whether Start, Build, or Scale applies.

Source check

Confirmed by: Anthropic's Claude Platform release notes confirm the rate-limit increase, the Sonnet/Haiku alignment with Opus tiers, and the Start/Build/Scale consolidation. Anthropic's rate-limit docs provide the stable tier reference.

Early signal / context: Releasebot independently indexed the same Claude Developer Platform entry and date. LinkLoot will treat console-visible limit changes or a pricing/rate-limit docs revision as update triggers.