The news
OpenAI released GPT-6 Astra to a limited set of organizations. The model will expand over the coming days to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.
The API pricing matches the rates set for Claude Fable 5 and 5.1. Input tokens cost $10 per million, while output tokens cost $50 per million.
Context
The release comes nearly two months after the general availability of the GPT-5.6 family of models. GPT-6 Astra is positioned as OpenAI’s latest frontier model and a direct competitor to Claude Fable 5 and 5.1.
Prior models from OpenAI followed a similar pattern of staged access before broader availability. The current rollout follows that approach without an immediate public launch for every user.
Simon Willison noted that the model is rolling out first to selected organizations. Full access for ChatGPT Plus, Pro, Business, and Enterprise accounts is scheduled to follow within days, along with availability on the OpenAI API and AWS. On OpenAI’s self-reported benchmarks, GPT-6 Astra scores higher than Claude Fable 5 and 5.1 on most tests. The pricing structure remains identical to the competing model’s rates: $10 per million input tokens and $50 per million output tokens.
Details
The rollout begins with a restricted group of organizations before reaching the full set of paid ChatGPT tiers. This sequence allows OpenAI to monitor load and performance while the model becomes accessible through both its own API and AWS. No independent evaluations appear in the initial announcements. The rollout timeline and pricing details constitute the primary concrete information released so far.
The model carries the same per-token costs as Claude Fable 5 and 5.1, removing price as a differentiator for teams already comparing the two providers. Input processing runs at $10 per million tokens and output generation at $50 per million tokens. Self-reported benchmark results from OpenAI place GPT-6 Astra ahead of the Fable versions on most internal tests, though these figures have not yet been replicated by outside reviewers.
The timing places GPT-6 Astra roughly eight weeks after GPT-5.6 reached general availability. This interval matches the pattern seen with earlier frontier releases, where OpenAI first opened the model to a narrow set of users before scaling. The phased method gives enterprise customers and API integrators an early window while keeping the model out of immediate reach for free-tier or individual accounts.
Why it matters
The staged release keeps the model out of immediate reach for most individual users while giving enterprise and API customers early access. This approach lets OpenAI manage capacity and gather feedback before scaling. For developers and companies already using the OpenAI API or AWS, the matching price point with Claude Fable 5 and 5.1 provides a direct comparison on cost. The self-reported benchmark gains may influence some procurement decisions, though external verification remains pending.
Teams that run workloads across multiple frontier providers now face identical token economics between GPT-6 Astra and the Fable line. Procurement conversations can therefore shift to measured performance, latency, and output quality rather than list price. The absence of third-party benchmarks at launch means initial decisions will rest on OpenAI’s internal numbers and whatever usage data the first organizations share.
The timing, two months after GPT-5.6 reached general availability, signals OpenAI’s intent to maintain a steady cadence of frontier releases. Organizations evaluating multiple providers now have another option at the same price tier, which could affect contract negotiations in the near term. Developers who have built pipelines around previous GPT releases can test the new model on existing infrastructure without changing billing structures. The limited initial distribution also reduces the risk of sudden capacity shortages that sometimes accompany wider launches.
---
Sources:
No comments yet