GPT-6 Astra is powerful, pricey and still gated

OpenAI introduced GPT-6 Astra last week as its most capable model for work, now live in ChatGPT Work, Codex and the API.

The pitch is that Astra does not require teams to rebuild data pipelines or integrations because it can operate the same applications people use, even when those applications do not have an API.

Early customers cited in the announcement include Cognition, Databricks and Hebbia, each reporting faster completion and cleaner artifacts on internal harnesses.

Silas Alberti, SVP Research at Cognition, said Astra delivered state-of-the-art performance on Devin's internal testing benchmark with noticeably easier-to-follow videos and clearer reports.

On paper the efficiency looks strong, with Astra hitting 57.9% on Terminal-Bench 4.0 versus 37.3% for GPT-5.6 Sol and 55.8% for Claude Fable 5.1 while costing about 9% and 63% less per task respectively.

Standard API pricing is $10 per million input tokens and $50 per million output tokens, with separate rates for cache reads and writes.

That headline rate is real and publicly listed, but the fine print matters because prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request.

Fast mode is available for GPT-6 Astra in the API and delivers up to 2x the speed of Standard processing at 2x the Standard price.

Safety claims rest largely on OpenAI's internal computer use safety benchmark, where the alignment table shows Astra at 2.4% versus 22.0% for GPT-5.6 Sol and 9.5% for Claude Fable 5.1.

The company also concedes that Astra's written reasoning is harder to monitor than GPT-5.6 Sol's and that extra safety checks can slow, pause or stop legitimate work in ChatGPT, Codex and the API.

Enterprise access is off by default at launch and requires administrators to enable Astra and configure allowed websites, desktop applications, uploads and confirmation policies.

Zero Data Retention is available only for eligible API customers subject to approval, and Astra is the first model to reach the Critical cybersecurity capability threshold under the Preparedness Framework, which brings tighter safeguards.

For tool-specific models like search and computer use there is a separate fee per tool call on top of tokens, so the token price alone does not determine total cost.

The verdict is that GPT-6 Astra API pricing is legit and not a scam, it appears good value for complex multi-step work where fewer tokens and retries offset the premium, but it is not yet proven cheaper at scale until independent enterprise results replace vendor benchmarks.