GPT-6 API: pricing and access

GPT-6 Standard text prices per million input/output tokens are Astra $10/$50, Sol $2/$10 and Luna $0.10/$0.50. OpenAI does not support Astra on its API Free tier. You can use OpenAI directly or a third-party provider; Kie advertises trial credits, while OpenRouter lists paid Astra access.

Compare GPT-6 API access options

Route	How to start	Price and trial status
OpenAI API	Create an API project and key, confirm model access, then use gpt-6-astra	Standard: $10 input / $50 output per million tokens; API Free tier is not supported
Kie AI	Select GPT-6 Astra in its Playground or create a Kie API key for gpt-6-astra	Kie advertises new-user trial credits; the reviewed page does not specify the amount, expiry or full eligibility conditions
OpenRouter	Create an OpenRouter API key and select openai/gpt-6-astra	Lists $10 input / $50 output per million tokens for Standard OpenAI routing; other provider and mode rates can differ

Each route uses its own account, key and billing. The request example below is for OpenAI directly. Kie and OpenRouter instructions and offers come from their own service pages, not from OpenAI. Provider details were checked on September 15, 2026; trial redemption was not tested.

Is there a free GPT-6 API?

There is no GPT-6 Astra Free tier in the OpenAI model reference. A ChatGPT subscription also does not include OpenAI API credit.

Kie says new users receive credits to try GPT-6 Astra, including its Playground and API offering. Check the credit balance and applicable terms before sending requests: the public page does not establish a specific free allowance or ongoing free use. OpenRouter’s Astra listing is paid; its separate free-model offerings do not make Astra free.

GPT 6 API model ID

The GPT-6 API model IDs are gpt-6-astra, gpt-6-sol and gpt-6-luna. Sol and Luna launched on September 22; Astra was the initial family model. Use the exact identifier shown in the OpenAI model reference and verify that your API organization can call it. GPT-6 Astra began limited rollout on September 3, 2026, with broader availability described as arriving over the following days. The existence of a public model page does not guarantee access for every new API account.

The model identifier and rollout context come from OpenAI’s GPT-6 Astra Model and GPT-6 Astra: A new generation of intelligence.

The API is a different access path from ChatGPT. Chat users select Astra as GPT-6 Pro; Sol and Luna are offered in Work and Codex, while developers authenticate an API project and send requests to an API endpoint. Billing, rate limits, tool support, data controls, and failure behavior can differ. ChatGPT 6 explains the product side; GPT 6 release date records the rollout language.

Choose a GPT-6 API model

Use Astra for demanding connected work, Sol for coding and agents, or Luna for focused high-volume tasks. All three publish a 1,050,000-token context and 128,000-token output limit. The family comparison tool compares your usage and required capabilities.

Use Responses for built-in tools and function calling. Sol and Luna support Chat Completions function calling only with reasoning effort none.

Current GPT 6 API pricing

These Astra rates come from its official model page. The family pricing guide also includes Sol and Luna from the shared configuration.

Token category	Standard rate
Uncached input	$10.00
Cached input	$1.00
Cache writes	$12.50
Output	$50.00

Requests above 272K input tokens use higher rates for the full request. Batch, Flex and Fast have their own pricing multipliers. Calculate your actual API usage cost or read the complete pricing guide for ordinary, cached and long-context estimates. Recheck the official rates before setting a customer price.

Budget for the whole job: instructions, retrieved documents, tool results, retries, evaluator calls and billable output, including reasoning tokens where applicable. Cache stable context when eligible, cap output and leave headroom for tool fees and repeat requests. A token unit price is not a per-user subscription estimate.

OpenAI API billing is separate from ChatGPT billing. Check the billing dashboard for your project’s organization before testing. Review current payment requirements and credit terms rather than inferring API funding from an active ChatGPT subscription. Record the billing organization alongside request results.

Endpoints and capabilities

The model reference lists Chat Completions and Responses, with Responses as the natural starting point for tool-using workflows. It lists support for streaming, function calling, structured outputs, web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, and MCP. The page also lists a 1,050,000-token context window and 128,000-token maximum output.

Capability flags are not a complete integration test. Verify the exact endpoint, SDK version, account permission, tool configuration, and data shape you intend to use. A workflow that works in a ChatGPT interface may depend on product-managed context or tools that your API request has not configured. A workflow that works for text may fail when a browser, shell, or external MCP server is unavailable.

Source: OpenAI API’s GPT-6 Astra Model.

A safe first GPT 6 API test

Start with a server-side request using a project-scoped API key. Never place the key in browser JavaScript, a public repository, or a client downloadable by users. Ask for a small structured response, record usage, and validate the returned fields. Keep the test input synthetic or low sensitivity until you understand retention, abuse monitoring, and your organization’s data settings.

Use three test cases: a normal request, an incomplete request that should return an explicit unknown, and a request that exercises the tool or structured format your product needs. Record latency, token counts, errors, and the model ID. Repeat the test after changing the prompt or SDK. This creates a baseline before you promise throughput or quality.

For production, add timeouts, retry rules, rate-limit handling, request tracing, spend limits, and a human review path for consequential actions. Store raw evidence separately from generated summaries. If your product sends customer data, document what is retained, who can see it, how deletion works, and whether zero data retention is available for your organization.

A minimal GPT 6 API request

This documentation-based shell example uses Sol through Responses. Change the model field to gpt-6-astra or gpt-6-luna to use another family member. Set OPENAI_API_KEY privately in your local environment first. Running it requires model access and incurs provider charges. Our resource experiments use Codex; this page does not report a paid API call.

curl --fail-with-body https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-sol","input":"Explain the difference between a draft and a verified result in one sentence.","max_output_tokens":512}'

Inspect the response status, output items, and usage fields. A 401 suggests an authentication problem. A 403 or 404 with a model-access error calls for checking the model ID and organization permissions. A 429 can indicate a rate limit or exhausted quota: inspect the error before retrying. For transient 5xx failures, use bounded retries with delay and keep the request identifier for support. Never publish the authorization header in a bug report.

Rate limits and rollout limits

The model page lists rate limits by usage tier, including requests per minute, tokens per minute, and batch queue limits. Your tier determines the actual ceiling. A product that works for one developer can fail under concurrent customer traffic if it assumes the highest tier. Measure the 95th-percentile job size, then reserve headroom for retries and bursts.

Rollout status is another limit. Availability and model permissions can vary by organization, product, and rollout stage. Confirm the current model list and run a real low-volume request. Do not advertise “unlimited GPT 6 API” when the underlying account has a finite rate limit, credit balance, or tool quota.

Source: OpenAI API’s GPT-6 Astra Model.

A developer acceptance checklist

Check authentication, model access, response parsing, the tools you need, rate-limit responses and timeout behavior. Keep keys on the server, cap input and output sizes, and record usage for a small representative request before estimating a customer budget. A retry should not repeat an external action such as sending a message or changing a record.

The prompt library contains a synthetic API-contract drafting example. It tests whether an interface description is internally consistent; it does not call the OpenAI API. For task framing and source checks, use the working guide.