GPT-6 Astra
OpenAI's GPT-6 flagship: $10 / $50 per million tokens, 1,050,000-token context, the only model with an Ultrafast tier.
What it is
GPT-6 Astra is the top of OpenAI's GPT-6 line, released September 3, 2026. It has no mini tier; the cheaper GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna sit below it. On September 29 OpenAI added an Ultrafast service tier for Astra at 6x the standard price.
Pricing
| Input | $10.00 |
| Output | $50.00 |
| Cached input (cache read) | $1.00 |
| Cache write | $12.50 |
| Batch input | $5.00 |
| Batch output | $25.00 |
| Fast / priority input | $20.00 |
| Fast / priority output | $100 |
- Long context (>272K input): $20 input, $75 output.
- Ultrafast tier: $60 input, $6 cached, $300 output (service_tier: ultrafast).
- Batch and Flex are 50% of Standard; Fast mode is 2x.
- Does not support the none reasoning effort level; no custom temperature or top_p.
Capability highlights
- OpenAI calls it its most intelligent and most aligned model (vendor claim).
- Ultrafast mode (from Sep 29, 2026) reduces time between output tokens; US data residency only.
- Async tool calling and mid-turn steering over WebSockets in the Responses API.
- Tool calling requires the Responses API.
Modalities: text, image → text. Reasoning model; effort low, medium, high, xhigh, max. Prompts over 272K input tokens are billed at long-context rates for the whole request.
How to use it
API: OpenAI API model ID gpt-6-astra (Responses, Chat Completions, Batch). Also offered through Microsoft Azure and AWS Bedrock per OpenAI's launch post.
Versus GPT-5.6 Sol
OpenAI's previous top tier listed at $4 / $20; Astra is a new flagship tier above it at 2.5x the price.
| GPT-5.6 Sol | Astra | |
|---|---|---|
| Input / 1M | $4.00 | $10.00 |
| Output / 1M | $20.00 | $50.00 |
| Cached input / 1M | $0.400 | $1.00 |
| Context window | not published | 1.05M |
| Max output | not published | 128K |
Published benchmark numbers
Vendor-reported, quoted from the linked announcement. Not independently verified; different vendors run benchmarks under different settings.
| Benchmark | Astra | Comparison in the same source | Source |
|---|---|---|---|
| Terminal-Bench 4.0 As published in Google's Gemini 4 Argon comparison table. | 58.2% | Gemini 4 Argon: 57.4%, Opus 5.5: 66.4% | Google DeepMind Gemini 4 Argon page (benchmark table) |
| DeepSWE v1.1 As published in Google's Gemini 4 Argon comparison table. | 74.1% | Gemini 4 Argon: 77.9%, Opus 5.5: 74.2% | Google DeepMind Gemini 4 Argon page (benchmark table) |
Comparisons
Sources
Every number above comes from one of these pages, read on the date shown. Searched for: astra, gpt 6 astra, gpt-6 astra.
- OpenAI model page: gpt-6-astra read 2026-10-08
context, max output, cutoff, pricing, features - OpenAI API pricing read 2026-10-08
standard, batch, flex, fast, ultrafast, long-context rates - OpenAI API changelog (Sep 3 and Sep 29, 2026 entries) read 2026-10-08
release date, Ultrafast mode - OpenAI announcement: GPT-6 Astra read 2026-10-08
availability; page returned 403 to our fetcher, availability taken from search snippet - Google DeepMind Gemini 4 Argon page (benchmark table) read 2026-10-08
third-party vendor-published scores