Free-tier rate limits tracker
Provider free tiers are metered on several axes at once and the numbers change with each model release. This page keeps the current values where a provider publishes them, links the source, and explains how each provider tells you which limit you hit. Last verified 2026-09-21.
Groq — Free plan
Per model. Groq returns 429 (rate over time) or 413 (single request over the per-minute budget); headers x-ratelimit-remaining-requests (daily) and x-ratelimit-remaining-tokens (per minute) are on every response. Developer-tier upgrades were paused ("temporarily unavailable due to high demand") at verification time.
| Model | RPM | RPD | TPM | TPD | Audio / hr | Audio / day |
|---|---|---|---|---|---|---|
openai/gpt-oss-120b | 30 | 1,000 | 8,000 | 200,000 | — | — |
openai/gpt-oss-20b | 30 | 1,000 | 8,000 | 200,000 | — | — |
qwen/qwen3.8-27b | 30 | 1,000 | 8,000 | 200,000 | — | — |
groq/compound | 30 | 250 | 70,000 | — | — | — |
groq/compound-mini | 30 | 250 | 70,000 | — | — | — |
whisper-large-v3-turbo | 20 | 2,000 | — | — | 7,200 s | 28,800 s |
whisper-large-v3 | 20 | 2,000 | — | — | 7,200 s | 28,800 s |
canopylabs/orpheus-v1-english | 10 | 100 | 1,200 | 3,600 | — | — |
Source: Groq rate limits. Entries: 429, 413, upgrade unavailable.
Google Gemini API — Free tier
Per model and per project: RPM, TPM and RPD. Google changes the numbers with each model release and publishes them in one table, so we link rather than copy them. The error is 429 with status: RESOURCE_EXHAUSTED; details[].quotaId names the window and retryDelay is set for per-minute quotas. Daily quotas reset at midnight Pacific. Attaching billing moves the project to Tier 1 immediately.
Source: Gemini rate limits. Entry: 429 RESOURCE_EXHAUSTED.
OpenAI — Usage tiers
OpenAI has no free API tier; limits scale with cumulative spend through Tiers 1–5, per model, on RPM, TPM and (for some models) requests per day. Two different 429s exist: rate_limit_exceeded (back off; headers x-ratelimit-remaining-* and retry-after) and insufficient_quota (the account is out of credit; retrying is pointless).
Source: OpenAI rate limits. Entry: 429 insufficient_quota, "You exceeded your current quota".
DeepSeek API — Concurrency cap
DeepSeek does not meter requests or tokens per minute. It caps requests in flight per account, across all API keys and all user_id values, and returns 429 with Rate Limit Reached above the cap. A request holds a slot from send until the last token, so long reasoning answers lower the effective throughput. Higher caps are granted on request at no charge. 503 Server Overloaded and 500 are separate: DeepSeek's capacity is full, most often in its peak-price hours.
| Model | Concurrent requests per account | Over the cap |
|---|---|---|
deepseek-flash | 2,500 | 429 Rate Limit Reached |
deepseek-v4-pro | 500 | 429 Rate Limit Reached |
Source: DeepSeek rate limit and isolation. Entry: 429 Rate Limit Reached / 503 Server Overloaded.
OpenRouter — Free models
Models with a :free suffix are limited per account by requests per minute and per day, with a higher daily allowance once the account has purchased credits. Paid models are limited by credit balance rather than by request counts. The error is 429; 402 means insufficient credits.
Source: OpenRouter limits.
Inference APIs
No free-tier caps and no tiers to unlock. Usage is limited by prepaid balance; requests return 402 insufficient_balance when it reaches zero. There is no per-minute or per-day request or token meter. Traffic that is abusive or endangers the service for other users may be throttled; see pricing.
Reading which limit you hit
| Provider | Where the answer is | Per-minute → wait | Per-day → do not wait |
|---|---|---|---|
| Groq | error message ("tokens per minute" / "requests per day"); headers | retry-after | x-ratelimit-reset-requests |
| Gemini | details[].violations[].quotaId | retryDelay | quotaId contains PerDay |
| OpenAI | error.code | rate_limit_exceeded + retry-after | insufficient_quota (add credit) |
| OpenRouter | HTTP status | 429 | 402 (add credits) |
| DeepSeek | message | Rate Limit Reached (concurrency; clears as requests finish) | Insufficient Balance (402, add funds) |
Values are copied from provider documentation on the verification date and re-checked when providers announce changes. If you see a different number in your console, tell us.