Billing and usage
How every Ostra request is priced and charged — Ostra Decision per-unit prices, the 24-hour cache, batch discount, Ostra Pro token pricing and up-front holds, spend limits, and the cost headers.
With an API key, Ostra is pay as you go and prepaid. You add credit in the TestRelic Console and each request is charged against the workspace balance as it runs. The Console has no subscription, no free tier and no invoice, and its credit is separate from a TestRelic platform plan. The Growth, Pro and Enterprise plans also include Ostra usage, metered and reset automatically; see pricing.
Prices
Ostra Decision
| Task | Billed per | Input (per 1M) | Batch input (per 1M) | Output |
|---|---|---|---|---|
| Failure triage | Failure | $0.50 | $0.25 | $0.00 |
| Test impact | Request (any number of candidates) | $2.00 | $1.00 | $0.00 |
| Selector survival | Locator | $0.25 | $0.125 | $0.00 |
- A single triage call costs $0.0000005; a million cost $0.50.
- The decisions it returns are free. A
deferanswer is billed like any other. - 24-hour cache: the same task with the same input (after normalising key order) within 24 hours, in the same workspace, is answered from cache at no cost (
cached: true,billed.units: 0). The cache resets when a new Ostra Decision release goes live. - Batch: each
/batchitem costs half its/predictprice, rounded up to the nano-dollar. - REST, MCP and the Console Playground cost the same.
Ostra Pro
| Per 1M tokens | |
|---|---|
| Input | $0.64 |
| Cached input | $0.06 |
| Output | $4.47 |
- Cost = (input − cached) × $0.64/1M + cached × $0.06/1M + output × $4.47/1M.
- Reasoning tokens bill as output. Lower
reasoning_effort, or turn thinking off, for cheaper answers. - Cached input is the part of the prompt served from the prompt cache. Follow-up turns of a conversation hit it most when you send
x-ostra-session.
When you're charged
Ostra Decision charges before the model runs and refunds the charge if the request fails. A batch is charged up front for all its items and refunded per failed item.
Ostra Pro can't know the cost in advance, so it holds the most a request could cost, then charges the real cost when it ends and releases the rest:
hold = min(262,144, prompt size + 16,384 per image + 65,536 per video) × $0.64/1M
+ max_tokens × $4.47/1MWith the default max_tokens (8,192) a text request holds at most about $0.20; the absolute maximum is about $0.31. If your balance can't cover the hold, the request fails with 402 insufficient_balance and its message tells you to lower max_tokens.
What's free
- Cached Ostra Decision answers.
- Any request that fails: validation errors, auth errors, 402, 429, model errors, timeouts, and Ostra Pro requests that time out while the model starts.
GET /models,POST /warm,ostra_list_models, MCPinitializeandtools/list.
An Ostra Pro stream that breaks after it started (your client disconnects, or the stream fails) is charged only for the tokens it produced. If the final usage never arrived, output is estimated from the text length and the log entry says "(estimated)".
Limits that stop spending
| Limit | Where | What happens |
|---|---|---|
| Balance | Billing | At $0, requests fail with 402 insufficient_balance. Turn on auto top-up to refill from a saved card. |
| Workspace monthly spend limit | Billing → Spent this month → Edit limit | Once this calendar month's (UTC) spend reaches it, requests fail with 402 spend_limit_reached, scope: "org", even with credit left. |
| Key monthly spend cap | API keys → Create / Edit | The same, for one key: scope: "key". |
When spend crosses the alert percentage of the workspace limit (80% by default), Ostra sends a billing.spend_alert webhook and writes an audit-log entry, once a month.
Cost in responses
| Where | What |
|---|---|
billed in /predict and /batch responses | units, costNano (integer nano-dollars) and costUsd (exact decimal string). |
ostra-cost-usd header | Ostra Pro, non-streaming responses: what the request cost. |
ostra-balance-remaining-usd header | The balance after the request (approximate when requests run in parallel). |
usage in Ostra Pro responses | Token counts; prompt_tokens_details.cached_tokens (OpenAI format) or cache_read_input_tokens (Anthropic format). |
MCP ostra_pro_chat | cost_usd and balance_usd in structuredContent. |
Long non-streaming Ostra Pro answers (over 45 seconds) start sending the response before the cost is known, so they come without the cost headers; read the request in Logs instead.
For daily breakdowns by model, task, channel and key, use Usage in the Console (CSV export available). See Usage and logs.
Rate limits
Ostra's requests-per-minute limits by usage level, what counts as a request, the rate-limit headers, and how to handle 429s.
Streaming
Stream Ostra Pro answers as server-sent events — the OpenAI and Anthropic event formats, usage chunks, keep-alive comments during cold starts, and errors inside a stream.