GPT-6 Astra

Developers can use it through the API to build software in programming languages including COBOL, Fortran, and Visual Basic.

Availability

Try our latest models through the API

Pricing model

API cost

Deployment

Global processing and US data residency; EU and other regional inference residency aren't supported

Privacy

US data residency available; EU and other regional inference residency aren't supported

Licence

Not established

• GPT-6 Astra can process more context than previous models to produce more structured outputs. ^[1]

  With GPT‑6 Astra, Harvey can bring more context into the drafting process and generate more structured outputs.

• Harvey reports substantial improvements in document formatting and context awareness when using GPT-6 Astra. ^[1]

  Compared with other models, Harvey has seen substantial improvements in document formatting and context awareness with GPT‑6 Astra, helping customers get more complete documents that better reflect the material behind them.

• GPT-6 Astra is being used by third-party legal application Harvey to draft and synthesize legal documents. ^[1]

  Harvey uses GPT‑6 Astra to help lawyers analyze, synthesize, and draft from the information that shapes a matter, including court information, law firm documents, case law research, and other sources of legal context.

• Because it can ingest more context, GPT-6 Astra enables applications to incorporate user preferences directly into drafting workflows.
• Using GPT-6 Astra, Harvey produces more complete documents that better reflect underlying legal material. ^[1]

  Compared with other models, Harvey has seen substantial improvements in document formatting and context awareness with GPT‑6 Astra, helping customers get more complete documents that better reflect the material behind them.

• Prompt caching reuses eligible shared prefixes within a 30-minute window to improve cache hit rates. ^[2]

  We now give cache discounts for eligible shared prefixes reused within a 30-minute window.

• Developers can receive discounts of up to 90% on cached input tokens. ^[2]

  OpenAI caches that shared context to reuse computation across requests, reducing response times and giving developers discounts of up to 90% on cached input tokens.

• A Prompt Caching Dashboard is available to show cache hit rates and input composition over time. ^[2]

  Track hit rates over time and use the input composition chart to compare cached and uncached tokens.

• A prompt caching diagnostics tool is available to identify reasons for cache misses and estimate affected tokens. ^[2]
• On GPT-6 models, reasoning effort can be changed between responses without breaking cache by using a configuration_update. ^[2]

  **Adjust reasoning effort without breaking cache.** On GPT‑6 models, you can now [change reasoning effort⁠(opens in a new window)](https://developers.openai.com/api/docs/guides/reasoning?api-mode=responses#change-reasoning-mid-conversation) between responses without breaking cache.

• Ultrafast mode reduces the time between generated output tokens. ^[3]

  Use gpt-6-astra with service_tier: "ultrafast" to reduce the time between generated output tokens.

• Ultrafast mode is available to API customers, subject to rate limits. ^[3]

  It is available to API customers, subject to rate limits, with global processing and US data residency.

• Ultrafast mode uses global processing and supports US data residency. ^[3]

  It is available to API customers, subject to rate limits, with global processing and US data residency.

• EU and other regional inference residency aren't supported. ^[3]

  EU and other regional inference residency aren't supported.

• GPT-6 Astra is available through the API.
• GPT-6 Astra achieves an average test pass rate of 87.4% across 20 programming languages at maximum reasoning effort. ^[4]

  At maximum reasoning effort, GPT‑6 Astra achieves an average test pass rate of 87.4% across the 20 languages, up from 26.8% for GPT‑5.6 Sol.

• GPT-6 Astra scores above 83% in 19 of the 20 languages tested. ^[4]

  Astra scores above 83% in 19 of the 20 languages, with COBOL remaining the hardest at 72%.

• GPT-6 Astra achieves a 72.1% test pass rate in COBOL. ^[4]

  On MultiLangBench, GPT‑6 Astra achieves an average test pass rate of 87.4% across 20 programming languages, up from 26.8% for GPT‑5.6 Sol, with improvements in every language tested.