Torki API
Usage Tiers and Rate Limits
Overview
Your organization's usage tier sets its rate limits and its monthly spend cap. The tier is set by the credit you have bought and rises automatically as you buy more.
Usage tiers
| Tier | How to get there | Monthly spend cap |
|---|---|---|
| Evaluation | New account, no purchase yet | $10 |
| Tier 1 | Any credit purchase | $100 |
| Tier 2 | $50 of credit bought in total | $500 |
| Tier 3 | $250 of credit bought in total | $2,000 |
| Custom | Contact sales@torkiai.com | Custom |
Your first purchase of any size moves you to Tier 1. Promotional credit doesn't count toward a tier.
Rate limits
Limits apply per organization, shared by all its workspaces and keys, and per model group: Pulse and Orbit share one set, Stellar and Cosmos another.
| Tier | Pulse and Orbit RPM · ITPM · OTPM | Stellar and Cosmos RPM · ITPM · OTPM | In flight |
|---|---|---|---|
| Evaluation | 20 · 50K · 8K | 10 · 30K · 20K | 2 |
| Tier 1 | 50 · 150K · 20K | 25 · 100K · 40K | 4 |
| Tier 2 | 150 · 400K · 50K | 75 · 250K · 80K | 8 |
| Tier 3 | 400 · 1M · 120K | 200 · 600K · 150K | 16 |
| Custom | Set for your organization | Set for your organization | 16 |
- RPM (requests per minute) is smoothed per second: after a short burst, about RPM ÷ 60 requests a second.
- ITPM (input tokens per minute) counts uncached input only.
- OTPM (output tokens per minute) counts the output actually produced, thinking included — not
max_tokens. A failed thinking attempt counts too, even when it isn't charged. - In flight is how many requests your organization may have running at once, all models together. A stream counts until it ends.
- Max prompt tokens: 131,072 per request, the context window, prompt plus output.
See your tier and every limit in Settings → Limits and tiers. Limit usage, under Analytics, shows how close you came to each limit in the last 24 hours and lists recent refusals.
When you hit a limit
Over a rate limit, the API returns 429 with a retry-after header in seconds, and the message names the limit that fired (your organization's, or a workspace's lower one). OpenAI SDKs retry these automatically.
429 rate_limit_exceeded— too many requests or tokens in a minute. Wait forretry-after, or spread requests out.429 concurrency_limit_exceeded— your organization already has its tier's number of requests in flight. Retry when one of them finishes.429 spend_cap_reached— your tier's monthly spend cap. There is noretry-after: it resets on the 1st at 00:00 UTC, or buy credit to move up a tier. Don't retry it in a loop.
Every response carries OpenAI-style headers, so you can slow down before you hit a limit: x-ratelimit-limit-requests, x-ratelimit-remaining-requests, x-ratelimit-reset-requests, x-ratelimit-limit-tokens, x-ratelimit-remaining-tokens and x-ratelimit-reset-tokens (tokens = input tokens), plus x-ratelimit-limit-output-tokens and x-ratelimit-remaining-output-tokens. The remaining values are for the current minute.
A spend limit you set yourself is not a rate limit: it returns 400 usage_limit_reached. Raise it in Settings → Billing, or in the workspace's Set limits. See API Credit, Billing and Invoices.
Workspace limits
An admin can give a workspace lower limits than the organization's: requests, input tokens and output tokens per minute for each model group, and a monthly spend limit. Open Workspaces, then the workspace's menu, then Set limits. Leave a field empty to use the organization's.
FAQ
How do I get higher limits?
Buy more credit: $50 in total reaches Tier 2 and $250 reaches Tier 3. For more than Tier 3, email sales@torkiai.com about a Custom tier.
Does a long thinking answer use up my output limit?
Yes. OTPM counts thinking tokens, so a long think on Stellar or Cosmos uses a large share of a minute's output. That is why Stellar and Cosmos have their own, larger OTPM.
Do cached tokens count toward ITPM?
No. ITPM counts uncached input only.