Claude rate limit error: what "API error: rate limit reached" means and how to fix it

The exact messages and what each means

• API Error: 429 {"type":"rate_limit_error", ...} / "rate limit reached" — you exceeded requests, tokens or concurrency for the window. Wait for retry-after.
• rate limit reached for <model> with a per-model note — limits are per model; switching tier with /model often works immediately.
• "You've hit your usage limit" / "limit resets at …" — not a rate limit: that is the Pro/Max subscription's rolling usage cap. Wait for the reset or use an API key.
• 529 overloaded_error — capacity, not you. Retry with backoff; the errors guide covers it.
• 400 … tool use concurrency — too many parallel tool calls in one turn; not a 429 but shows up in the same sessions.

Claude rate limits: the three ceilings

Anthropic enforces requests per minute (RPM), input and output tokens per minute (ITPM/OTPM) and concurrent in-flight requests, per model, per organisation, scaled by usage tier — the tier rises automatically as your cumulative spend grows. Gateways enforce their own equivalents per account. The numbers are published in Anthropic's docs and shown in your console; the ones that matter for Claude Code are concurrency and OTPM, because an agent session is a stream of long, parallel, output-heavy requests.

Why Claude Code hits the limit when a chat client doesn't

A chat client sends one request at a time. Claude Code streams every turn (holding a concurrency slot for its whole duration), runs subagents in parallel (each one another slot), and resends the entire conversation on every turn (tokens per minute climb with session length). Ten subagents on a large repo is ten simultaneous streams and hundreds of thousands of tokens a minute — that is the 429.

Fixes, in order of how often they work

• 1. Let it retry. Claude Code honours retry-after; most 429s clear in seconds.
• 2. Run fewer subagents in parallel; sequence heavy tasks instead of fanning out.
• 3. /compact — smaller context means fewer tokens per turn, which is the TPM lever.
• 4. Switch tier with /model for the current task; limits are per model.
• 5. Move background work to Haiku; keep Sonnet/Opus for the main loop.
• 6. Ask for a higher limit: on Anthropic, spend tier or a request to sales; on this gateway, e-mail support with your account and the workload — new accounts start conservative on concurrency precisely so one runaway agent cannot exhaust the pool, and are raised on request.

/compact focus on the current task
/model claude-sonnet-5
# in scripts: cap the loop and let retries happen
claude -p "..." --max-turns 20

Rate limits on this gateway

Requests are billed against your balance or a flat-rate plan; rate limits are per account and independent of price. The gateway pools several upstream accounts, so a per-model 429 on one is failed over rather than surfaced when another has room — one reason sessions here see fewer 429s than a single console key. Concurrency is the limit you are most likely to meet; write to support with the number of parallel agents you run and it is adjusted.

Plan	Duration	Price
Unlimited 1 Hour	1 hour	$1
Unlimited 24 Hours	24 hours	$10
Unlimited 1 Week	1 week	$49
Unlimited 15 Days	15 days	$89

All unlimited plans →

Frequently asked questions

AI Prime Tech is an independent API gateway and is not affiliated with, endorsed by, or sponsored by Anthropic. “Claude” and “Claude Code” are trademarks of Anthropic. Claude Code features described here follow Anthropic’s public documentation at the time of writing and change frequently; prices and model lists on this page are read from this gateway’s live settings.