GPT-6 Astra: API pricing, specs and Codex limits

GPT-6 Astra at a glance

OpenAI calls GPT-6 Astra (gpt-6-astra) "our most capable model for the most demanding work," built to handle complex reasoning, coding, computer use, research, and document creation. Its knowledge cutoff is April 30, 2026, and the main specifications are in the table below.

GPT-6 Astra API pricing

OpenAI bills gpt-6-astra per token, using the Standard tier as the base rate and pricing Batch, Flex, Fast, and Ultrafast as multiples of that base.

On the Standard tier, a request with 200,000 input tokens and 20,000 output tokens costs $2.00 for input (200,000 x $10 / 1M) plus $1.00 for output (20,000 x $50 / 1M), or $3.00 in total. Running that same 200,000-input and 20,000-output job through the Batch API cuts the rate in half to $1.00 for input (200,000 x $5 / 1M) and $0.50 for output (20,000 x $25 / 1M), coming out to $1.50 total.

Cached input on Standard is $1 per million tokens, a tenth of the standard input price, while cache writes cost $12.50 per million tokens for a 30-minute cache and $25 per million for a 12-hour cache (on Batch and Flex, cached input drops to $0.50 per million, and on Fast it's $2). OpenAI also added Ultrafast mode for GPT-6 Astra in the Responses API on September 29, 2026, priced at $60 per million input tokens, $6 per million cached input tokens, and $300 per million output tokens; see our OpenAI API pricing page for the rest of the lineup.

What GPT-6 Astra supports

• The model accepts text and image inputs and returns text output only, with no support for audio or video.
• It supports streaming, function calling and structured outputs and runs on both Chat Completions and the Responses API, though fine-tuning isn't supported.
• Reasoning effort can be set to low, medium, high, xhigh, or max, and the model doesn't support the none effort level.
• You can't pass custom temperature or top_p values to gpt-6-astra, and it doesn't return log probabilities (logprobs).
• On OpenAI's API, tool calling with GPT-6 Astra requires the Responses API.
• On September 3, 2026, the Responses API added controls for long-running Astra work, including async tool calling, mid-turn steering over WebSockets, and the ability to change reasoning effort mid-conversation while keeping the cached prompt prefix.

GPT-6 Astra in Codex

If you're using GPT-6 Astra through a ChatGPT subscription instead of paying per token on the API, OpenAI includes it in Codex as "our most capable model for complex work across code, apps, and research," with usage limits that depend on your plan tier. The table below breaks down the message estimates and plan rules, and we cover the full system in our guide to Codex usage limits.

• Astra burns through Codex credits 5x faster than GPT-6.1 Sol on uncached input and output, at 250 credits per million input tokens and 1,250 per million output versus 50 and 250 on Sol (cached input is 25 credits per million on Astra versus 2.5 on Sol).
• OpenAI suggests starting Astra at Light effort (low in configuration), and some paid plans omit Astra Extra High, while Ultrafast in Codex uses included usage at 8x (6x on credits) on Pro 500 and eligible Enterprise and Edu plans, and Fast mode uses it at 2.5x.
• To launch it in the CLI, run codex -m gpt-6-astra.

GPT-6 Astra or GPT-6.1 Sol

OpenAI positions GPT-6 Astra for the hardest end-to-end work across code, apps, and research. For repeated, long-running work, OpenAI released GPT-6.1 Sol (gpt-6.1-sol) on September 29, 2026, pitching it as near-Astra performance for complex work at a lower cost; you can compare the rest of the family on our GPT-6 models page.

On the API, gpt-6.1-sol costs $2 per million input tokens and $10 per million output tokens for prompts up to 272K, which is a fifth of Astra's $10 and $50 rates (with cached input at $0.10 per million on Sol versus $1 on Astra). In Codex, a ChatGPT Plus ($20 a month) or Standard Business ($25 a seat monthly, or $20 billed yearly) account gets an estimated 15 to 160 local messages per five hours with GPT-6.1 Sol, compared to 5 to 45 with GPT-6 Astra.

Call GPT-6 Astra from your ChatGPT plan

If you already pay for a ChatGPT plan that includes Astra in Codex, you can use it as an API through your ChatGPT subscription with Plan2API. You sign up with a Google account at plan2api.com/connect and link your ChatGPT account using a one-time device code, and Plan2API reads the Codex models on your account live and exposes them as openai/<model>. Any plan with Astra in Codex can call openai/gpt-6-astra (codex/gpt-6-astra works too) in standard OpenAI Chat Completions format (POST /v1/chat/completions) at https://api.plan2api.com/v1 using your p2a_ key in an Authorization: Bearer header, with streaming, tool and function calling, JSON mode (response_format set to json_object or json_schema), image inputs, and reasoning_effort (reasoning.effort OpenRouter style works too) all on the same endpoint.

Requests draw on your ChatGPT plan's Codex allowance instead of running up a per-token API bill, while temperature and top_p are ignored on Codex models. Plan2API's Free plan ($0, no card needed) covers 1 connected subscription, 2 API keys, and 50 chat requests a day (resetting at 00:00 UTC), while Pro is $12 a month (or $99 a year) for 3 connected subscriptions, 10 API keys, and no daily request cap.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.plan2api.com/v1",
    api_key=os.environ["PLAN2API_KEY"],
)

response = client.chat.completions.create(
    model="openai/gpt-6-astra",
    reasoning_effort="high",
    messages=[{"role": "user", "content": "Plan the move of this service to a queue-based design."}],
)
print(response.choices[0].message.content)