OpenAI GPT-6 Astra API Goes Live, With Pricing and Rollout Date Discrepancies

On September 3, 2026, OpenAI released the GPT-6 Astra model with the API model ID gpt-6-astra. The context window reaches 1.05M tokens, with pricing of $10 per million input tokens and $50 per million output tokens, a maximum output of 128K tokens, and support for text and image inputs.

The Facts

According to multiple independent sources, OpenAI officially launched GPT-6 Astra on September 3, 2026, with the API ID confirmed as gpt-6-astra. Initial access was limited to specific organizations, then expanded to the OpenAI API and paid ChatGPT plans starting September 4. Pricing details include cached input at $1 per million tokens, batch processing at half price, and Fast mode at twice the standard rate. The knowledge cutoff date is April 30, 2026.

Mechanism Breakdown

Sources indicate that GPT-6 Astra is positioned for computer use and software engineering rather than pure conversation. OpenAI says the model completes tasks 1.9 times as fast as Sol on the Mind2Web benchmark. On cybersecurity, it scored 100% on ExploitBench, above Sol's 78.5%. On hallucination benchmarks it scored 4.2%, below Sol's 12.2%. A companion Codex update introduces note-taking across context windows, currently in the experimental stage.

The rollout follows a phased strategy: September 3 covered a limited set of organizations, and September 4 expanded to the API and paid plans. Enterprise access is disabled by default and must be enabled by administrators. Reports confirm the model meets the "Critical" cybersecurity threshold of OpenAI's Preparedness Framework; the public version declines advanced cyber tasks, and relaxed safeguards apply only to verified organizations under the Daybreak program.

Industry Impact

For developers, gpt-6-astra offers a 1.05M context window with image input, making it directly usable for complex software engineering tasks, though teams will need to adapt to the phased access restrictions. Enterprise users can call the model through existing API keys at some gateways for 10% below official pricing, reducing integration costs.

On the competitive landscape, sources note that Qwen, GLM, DeepSeek, and Kimi have already launched compatible endpoints on the Yotta AI Gateway, with Claude currently being added. Adding Astra to the routing layer requires only a configuration change, but its pricing — 2.5 times that of GPT-5.6 Sol — may affect cost-sensitive deployments.

Comparisons and Precedents

Available information indicates that GPT-6 Astra evolved from the GPT-5.6 series, with significant architectural differences. In similar past launches, API availability and ChatGPT availability have often been out of sync, and this release again reflects that in the gap between limited organizational access on September 3 and the broader expansion on September 4.

Strategic Assessment

Based on available facts, the most likely near-term scenario is that more gateways will progressively open routing for gpt-6-astra, while independent benchmark results are released to verify OpenAI's self-reported figures such as the 57.9% score on Terminal-Bench 4.0. Signals to watch include the timing of official listings on Azure Foundry and Amazon Bedrock, as well as the expansion scope of trusted access programs.