Introduction
GPT 6 Astra handles difficult analysis, substantial code changes and professional documents through a text conversation that can also include images. Its 1,050,000-token context provides room for the materials surrounding a question, while the answer can use up to 128,000 tokens, including reasoning. Choose Astra when you can define what a satisfactory result must demonstrate and have time to inspect its evidence.
Key features & highlights
Astra offers low, medium, high, xhigh and max reasoning effort. There is no none setting. Increase effort for a problem that needs more careful exploration, rather than treating the highest setting as a substitute for clear instructions. Keep enough output budget for both the reasoning and the explanation you expect to receive.
Function definitions let Astra propose actions inside your application. Your application still validates arguments, decides whether an action is allowed, executes it and returns the result with the matching call identifier. Supplying a function does not grant the model access to a browser, filesystem or external account by itself.
Request parameters
| Parameter | Required | Description |
|---|---|---|
messages | Yes | Give Astra the source code or document evidence and the criteria used to judge its findings. Images may accompany text. at least 1 items · at most 256 items |
max_tokens | No | Astra shares this output allowance between reasoning and its visible answer; default 4,096, maximum 128,000. |
reasoning_effort | No | Choose low, medium, high, xhigh or max for Astra; medium is the default, and none is unavailable. |
tools | No | Describe the application functions Astra may request; your application checks arguments and executes each call. at most 128 items |
parallel_tool_calls | No | Allow Astra to propose several independent function calls together; execute them only when your application permits concurrency. Default: |
tool_choice | No | For Astra, choose auto, none, required or a named function according to the allowed response contract. |
response_format | No | Use a supported text, JSON object or JSON schema response format for Astra and validate its returned values. |
Pricing
Input $8 / Output $40 per 1 million tokens; Prompts over 272K tokens: Input $16 / Output $60
Use cases
For a difficult migration review, include the old contract, the proposed schema and the rollback conditions. Ask Astra to produce a table of incompatible assumptions, the affected call sites and a confirming check for each one. A useful review distinguishes a demonstrated incompatibility from a possibility that still needs a test.
For a policy analysis, label the source documents and their effective dates. Request findings linked to those labels, followed by unresolved questions. Keep the original passages available to the reviewer so an apparently confident conclusion can be checked against the controlling text.
Prompt guide & pro tips
State the decision to support, the constraints that cannot change and the required deliverable before asking Astra to work. Mark quoted material as evidence rather than instructions. For a patch request, include the failure and the expected behavior; for an analysis, specify the evidence needed to justify a conclusion.
Use structured output when a downstream program requires particular fields, and validate the returned data before acting on it. Stream long replies when the client needs progress, but wait for the terminal event and measured usage before treating a run as complete. Requests above 272,000 input tokens enter the long-prompt pricing tier, so include extensive source material when it contributes to the question.
API quick start
curl --fail-with-body -sN https://api.spicyapi.ai/v1/chat/completions \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $(uuidgen)" \
-d '{
"model": "openai/gpt-6-astra/chat",
"messages": [
{
"role": "user",
"content": "Summarise the following release notes into three bullet points."
}
],
"stream": true
}' \
| jq --unbuffered -Rrj 'select(startswith("data: ")) | .[6:] | select(. != "[DONE]") | fromjson | if .error then "\n[error] \(.error.message)\n" else (.choices[0].delta.content? // empty) end'
See the API reference for authentication, the complete schema, polling, and error handling.

















