Post-Cutoff
Search For AI agents

OpenAIGPT-6 familyCurrentReasoning modelReleased 3 Sep 2026

GPT-6 Astra

Hardest end-to-end work: long-horizon agentic coding, research, analysis, computer use.

Model idgpt-6-astra

Price$10 in, $50 out per 1M tokens

Checked29 September 2026

Knows up to April 2026

1,863events after its cutoff

83 historic, 426 major. 197 days of news.

It launched on 3 September 2026, already missing 271 of them.

Five things it has never heard of

How to call it

4 ways to use GPT-6 Astra, with the model id each provider expects.

Providers, model ids, endpoints and docs for GPT-6 Astra
ProviderModel idEndpointDocs
OpenAI API gpt-6-astra https://api.openai.com/v1/responses developers.openai.com
Azure OpenAI (Microsoft Foundry) gpt-6-astra learn.microsoft.com
OpenRouter openai/gpt-6-astra openrouter.ai
Web app chatgpt.com

Pricing

Prices for GPT-6 Astra
Input$10 per 1M tokens
Cached input$1 per 1M tokens
Output$50 per 1M tokens
Cache write$12.50 per 1M tokens

As published: “per 1M tokens (USD), standard tier”. Note: prompts >272K input tokens billed at 2x input and cache rates.

Source: developers.openai.com, checked 29 September 2026.

Specs

Input
Text and images
Output
Text
Context window
1,050,000 tokens
Max output
128,000 tokens
Open weights
No
Licence
Proprietary
Training cutoff
April 2026
Released
3 September 2026
Status
Current
Type
Reasoning model

What stands out

What its maker marketed at launch, and what people found later, each with its source.

  • Max reasoning effort

    reasoning.effort adds a new “max” level above xhigh (low/medium/high/xhigh/max).

    Source: developers.openai.com

  • 1M-token context

    1.05M context window (922K max input) with 128K output on the flagship.

    Source: developers.openai.com

  • Restricted cyber behaviour

    Released as a restricted version that rejects certain cybersecurity prompts; separate Cyber/Daybreak models exist for that domain.

    Source: en.wikipedia.org

  • Recurrent-depth reasoning

    Found after launch

    Reported new “recurrent depth” technique that obscures some of the reasoning, raising monitorability concerns among safety researchers.

    Source: en.wikipedia.org

  • Ultrafast service tier

    Found after launch

    From Sept 29 2026, service_tier “ultrafast” gives up to 6x faster generation in the API (8x / ~300 tok/s in Codex) at 6x price: $60 input / $6 cached / $75 cache write / $300 output per 1M tokens (<=272K ctx); default limits 500K-5M TPM.

    Source: developers.openai.com

  • Powers dots always-on agents

    Found after launch

    OpenAI’s dots (launched Sept 29 2026) run on GPT-6 Astra, each with its own cloud computer; also the default model in Agents API computer-use examples.

    Source: openai.com

Four assistants, one scale

Each row runs from what the model knows (graphite) through the months between its cutoff and its launch (hatched) to today (pink).

  1. GPT-6 Astra 271 at launch, 1,863 today
  2. Claude Opus 5.5 372 at launch, 1,795 today
  3. Gemini 4 Hydrogen 1,863 at launch, 1,889 today
  4. Grok 4.7 387 at launch, 1,829 today
Show as a table
ModelCutoffReleasedMissing at launchMissing today
GPT-6 AstraApril 20263 September 20262711,863
Claude Opus 5.5June 202622 September 20263721,795
Gemini 4 HydrogenMarch 202612 November 20261,8631,889
Grok 4.7May 202621 September 20263871,829

Notes

OpenAI flagship (“most capable model, built for the hardest end-to-end work”). API changelog: Sep 3 2026 (limited preview Sep 3, public Sep 4). Single snapshot gpt-6-astra. OpenRouter also lists openai/gpt-6-astra-pro = same model with reasoning.mode pro. Endpoints: Chat Completions, Responses, Batch.

curl https://api.openai.com/v1/responses \
 -H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" \
 -d '{"model":"gpt-6-astra","input":"Hello"}'

Sources: