Elevenlabs Rate Limits

ElevenLabs throttles on CONCURRENCY, not on a requests-per-minute quota. The docs say so explicitly: "API requests per minute and concurrent requests are different metrics" and no per-minute ceiling is published. The limit that binds is how many requests a plan may have in flight at once, and it differs by model family (Multilingual v2 vs Flash), by Speech to Text (elevated), by realtime STT and by Music. Beyond the ceiling requests are queued rather than rejected — the docs put the added latency at roughly 50ms — and a 429 is returned when the queue itself is exceeded.