Alex Getmanearly AI signals
EN

GPT-6 Astra

OpenAI

Released Updated
Context window
1M
1,050,000 tokens
Max output
128K
128,000 tokens
Input / Output
text, image → text

API pricing

USD / 1M tokens

InputOutputCached input
  • $10.00
openrouter
  • $50.00
openrouter
  • $1.00
openrouter
Calculate your workload

Using GPT-6 Astra for a demanding task

Sources checked:

OpenAI's model reference positions Astra for demanding reasoning, coding and research. For a useful trial, choose a deliverable with a checkable result: a repository change with tests, an analysis whose claims can be traced to documents, or a structured extraction you can validate.

Watch the full request size. The reference describes a different price tier above 272,000 input tokens. Read the conditions beside this page's prices before budgeting a large document or a long agent session. A context-window maximum is a capacity limit, not a recommendation to send every file. Include tool calls, cache writes, retries and human review in the API task-cost worksheet.

Keep the model and the tool separate. A trial in Codex also tests its instructions, tools and permissions. Record those settings alongside the model and reasoning effort. The Codex guide and coding-tool comparison explain how to compare the workflow. These are suggested checks, not results of a new benchmark.

Timeline

  1. · API availability · evidence link · evidence link
  2. · API availability · evidence link
Input / output price · USD / 1M tokens
ModelEvidence linksInput / output price
gpt-6-astra
batch · long context
openai
API route unavailable
$10.00 (Input > 272,000 tokens) / $37.50 (Input > 272,000 tokens)
gpt-6-astra
batch · short context
openai
API route unavailable
$5.00 (Input ≤ 272,000 tokens) / $25.00 (Input ≤ 272,000 tokens)
gpt-6-astra
fast · long context
openai
API route unavailable
$40.00 (Input > 272,000 tokens) / $150.00 (Input > 272,000 tokens)
gpt-6-astra
fast · short context
openai
API route unavailable
$20.00 (Input ≤ 272,000 tokens) / $100.00 (Input ≤ 272,000 tokens)
gpt-6-astra
flex · long context
openai
API route unavailable
$10.00 (Input > 272,000 tokens) / $37.50 (Input > 272,000 tokens)
gpt-6-astra
flex · short context
openai
API route unavailable
$5.00 (Input ≤ 272,000 tokens) / $25.00 (Input ≤ 272,000 tokens)
gpt-6-astra
standard · long context
openai
API route unavailable
$20.00 (Input > 272,000 tokens) / $75.00 (Input > 272,000 tokens)
gpt-6-astra
standard · short context
openai
API route unavailable
$10.00 (Input ≤ 272,000 tokens) / $50.00 (Input ≤ 272,000 tokens)
gpt-6-astra
ultrafast · long context
openai
API route unavailable
$120.00 (Input > 272,000 tokens) / $450.00 (Input > 272,000 tokens)
gpt-6-astra
ultrafast · short context
openai
API route unavailable
$60.00 (Input ≤ 272,000 tokens) / $300.00 (Input ≤ 272,000 tokens)
openai/gpt-6-astra-pro
standard
openrouter$10.00 / $50.00
openai/gpt-6-astra-pro:batch
batch
openrouter$5.00 / $25.00
  1. · Confirmed use · OpenAI · official articles · How to connect AI usage to business value
    Excerpt
    Playco⁠ uses GPT‑6 Astra through OpenAI’s API to build and test playable game prototypes
  2. · Confirmed use · OpenAI · official articles · Harvey turns legal context into stronger drafts with GPT-6 Astra
  3. · Confirmed use · OpenAI · official articles · A model guide for the GPT-6 family
  4. · Evidence links · OpenAI · official articles
    Excerpt
    When we introduced GPT‑6 Astra, we demonstrated how far our models have come in using computers for professional work, from preparing documents to testing websites
  5. · Confirmed use · OpenAI · official articles · How Jump Trading is scaling quant research with ChatGPT
    Excerpt
    Jump Trading uses GPT‑6 Astra to take on longer, more ambiguous research problems
  6. · Evidence links · OpenAI · official articles
    Excerpt
    Jump Trading uses GPT‑6 Astra to take on longer, more ambiguous research problems
  7. · Evidence links · OpenAI · official articles
  8. · Evidence links · OpenAI · official articles
    Excerpt
    Making longer videos possible with GPT‑6 Astra
  9. · Confirmed use · OpenAI · official articles · Asana cuts model costs 76x in browser tests with GPT-6.1 Sol
  10. · Evidence links · OpenAI · official articles
    Excerpt
    We used some of our most capable and aligned models, including our upcoming GPT‑6 Astra Cyber model and an internal frontier model
  11. · Confirmed use · OpenAI · official articles · Greptile takes bug detection beyond code review with GPT-6 Astra
    Excerpt
    How Greptile uses GPT‑6 Astra to test user workflows, Daybreak to strengthen security, and Codex to reduce rework
  12. · Confirmed use · OpenAI · official articles · Retell uses OpenAI models to help power 100M monthly calls
  13. · Confirmed use · OpenAI · official articles · Qodo builds context into AI code review with Codex
  14. · Confirmed use · OpenAI · official articles · DevRev delivers autonomous self-service with GPT-6.1 Sol
    Excerpt
    DevRev’s sales engineers also use GPT‑6 Astra to create demos for the company’s booths at international conferences
  15. · Confirmed use · OpenAI · official articles · Merge builds product and business tools with OpenAI technology
    Excerpt
    For its own product development, Merge uses GPT‑6 Astra
  16. · Evidence links · OpenAI · official articles
  17. · Evidence links · OpenAI · official articles
  18. · Confirmed use · OpenAI · official articles · Introducing Astra for Law
  19. · Evidence links · OpenAI · official articles
    Excerpt
    It will appear in the model picker as “GPT‑6 Astra Law” and in the API as gpt-6-astra-law
  20. · Evidence links · OpenAI · official articles
    Excerpt
    We’ve also brought GPT‑6 Astra’s improved communication style to Sol and Luna, which we think will be especially noticeable in technical and coding conversations
  21. · Catalogue listing · GitHub · openai/openai-node · code
    Excerpt
    ils and * recommended replacements. * - "gpt-3.5-turbo-1106": Announced shutdown date: 2026-09-28. See * https://developers.openai.com/api/docs/deprecations for details and * recommended replacements. */ model: | 'gpt-6-astra' | 'gpt-6.1-sol' | 'gpt-6.1-terra' | 'gpt-6-sol' | 'gpt-6-luna' | 'gpt-5.6-sol' | 'gpt-5.6-terra' | 'gpt-5.6-luna' | 'gpt-5.5' | 'gpt-5.5-2026-04-23' | 'gpt-5.4' | 'gpt-5.4-mini' | 'gpt-5.4-nano' | 'gpt-5.4-mini-2026-03-17' | 'gpt-5.4-nano-2026-03-17' | 'gpt-5.3-chat-latest' | 'gpt-5.2' | 'gpt-5.2-2025-12-11' | 'gpt-5.2-chat-latest' | 'gpt-5.2-pro' | 'gpt-5.2-pro-2025-12-11' | 'gpt-5.1' | 'gpt-5.1-2025-11-13' | 'gpt-5.1-codex' | 'gpt-5.1-mini' | 'gpt-5.1-chat-latest' | 'gpt-5' | 'gpt-5-mini' | 'gpt-5-nano' | 'gpt-5-20
  22. · Confirmed use · OpenAI · official articles · Cogent and OpenAI build cyber defense factories for enterprises
  23. · Catalogue listing · GitHub · openai/codex · code
    Excerpt
    es::start_mock_server().await; let board = responses::start_mock_server().await; let url = board.uri(); let root = test_codex() .with_model("gpt-6-astra") .with_config(move |config| { configure(config); config.ephemeral = true; for feature in [ Feature::CodeMode, Feature::CodeModeOnly, Feature::CodeModeHost, ] { config .features

    This code reference does not establish a working model identifier.

  24. · Catalogue listing · GitHub · openai/codex · code
    Excerpt
    EL_ID: &str = "gpt-6.1-sol"; const GPT_6_1_TERRA_OPENAI_MODEL_ID: &str = "gpt-6.1-terra"; const GPT_6_SOL_OPENAI_MODEL_ID: &str = "gpt-6-sol"; const GPT_6_LUNA_OPENAI_MODEL_ID: &str = "gpt-6-luna"; const GPT_6_ASTRA_OPENAI_MODEL_ID: &str = "gpt-6-astra"; const GPT_5_5_OPENAI_MODEL_ID: &str = "gpt-5.5"; pub(crate) fn static_model_catalog() -> ModelsResponse { let catalog = ModelsResponse { models: vec![ bedrock_model( bundled_openai_model(GPT_6_1_SOL_OPENAI_MODEL_ID), AMAZON_BEDROCK_GPT_6_1_SOL_MODEL_ID, "GPT-6.1 Sol", /*priority*/ 0, ), bedrock_model( bundled_openai_model(GPT_6_ASTRA_OPENAI_MODEL_ID), AMAZON_BEDROCK_GPT_6_ASTRA_MODEL_ID, "GPT-6-Astra", /*priority*/ 1, ), bedrock_mod
  25. · Catalogue listing · GitHub · openai/openai-agents-python · code
    Excerpt
    ody={"reasoning_effort": "low"}), False, "low", ), ("gpt-6-sol", "string", "none", ModelSettings(), False, None), ("gpt-6.1-sol", "string", "function", ModelSettings(), False, None), ("gpt-6-astra", "string", "function", ModelSettings(), False, None), ("gpt-6-sol-unknown", "string", "function", ModelSettings(), False, None), ("gpt-6-sol", "string", "function", ModelSettings(), True, None), ], ) @pytest.mark.parametrize("stream", [False, True]) async def test_gpt6_tool_request_defaults( name: str, entry: str, tools: str, settings: ModelSettings, responses: bool, expected: str | None, stream: bool, ) -> None: if responses and stream: pytest.skip("Responses preservation is covered by the non-streaming request.") before = settings.to_json_dict() bodies: list[dic

    This code reference does not establish a working model identifier.

  26. · Catalogue listing · GitHub · openai/openai-python · code
    Excerpt
    ormat, and querying for objects via API or the dashboard. Keys are strings with a maximum length of 64 characters. Values are strings with a maximum length of 512 characters. """ model: Union[ Literal[ "gpt-6-astra", "gpt-6.1-sol", "gpt-6.1-terra", "gpt-6-sol", "gpt-6-luna", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna", "gpt-5.5", "gpt-5.5-2026-04-23", "gpt-5.4", "gpt-5.4-mini", "gpt-5.4-nano", "gpt-5.4-mini-2026-03-17", "gpt-5.4-nano-2026-03-17", "gpt-5.3-chat-latest", "gpt-5.2", "gpt-5.2-2025-12-11", "gpt-5.2-chat-latest", "gpt-5.2-pro", "gpt-5.2-pro-2025-12-11", "gpt-5.1", "gpt-
  27. · Catalogue listing · GitHub · openai/openai-agents-js · code
    Excerpt
    (): boolean { return isGpt5OrNewerReasoningModel(getDefaultModel()); } const DEFAULT_REASONING_EFFORT_PATTERNS: Array< readonly [ RegExp, Exclude<ModelSettingsReasoningEffort, 'minimal' | 'xhigh' | 'max' | null>, ] > = [ [/^gpt-6-astra$/, 'low'], [/^gpt-5(?:-\d{4}-\d{2}-\d{2})?$/, 'low'], [/^gpt-5\.1(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.2(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.2-pro(?:-\d{4}-\d{2}-\d{2})?$/, 'medium'], [/^gpt-5\.2-codex$/, 'low'], [/^gpt-5\.3-codex$/, 'none'], [/^gpt-5\.4(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.4-pro(?:-\d{4}-\d{2}-\d{2})?$/, 'medium'], [/^gpt-5\.4-mini(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.4-nano(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.5(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.6$/, 'none'], [/^gpt-5\.6-sol(?:-\d{4}-\d{2}-\d{2})?$/, 'none'], [/^gpt-5\.6-terra(?:-\d{4
  28. · Catalogue listing · Vertex AI · quotas
  29. · Alex's posts · Alex Getman
    Excerpt
    Kimi K3.1 Pro benchmarks are out! Of all these benchmarks, I trust Terminal-Bench 4.0 the most. And the results are impressive: Kimi K3.1 Pro scores 65.7, just 0.7 points behind the leader, Opus 5.5, while outperforming GPT-6 Astra by 7.5 points.
  30. · Alex's posts · model family · Alex Getman
    Excerpt
    Looks like Astra 6.1 is coming OpenAI rolled out a private test model gpt-6-astra-api-priv-ev3
  31. · Evidence links · OpenAI · news
    Excerpt
    Higgsfield brings creators longer videos with GPT-6 Astra, precise visuals with GPT-Image-2.5, and a path to voice filmmaking with GPT-Live
  32. · Alex's posts · Alex Getman
    Excerpt
    Meta just released Muse Wonder According to the benchmarks, it's right up there with GPT 6 Astra and Claude Opus 5.5. The results look impressive, but the model might be heavily optimized for benchmarks. It's still unclear how well it actually performs in real world use. And yeah, most of you probably won't be able to try it through a subscription, since Muse is still only available in the US.
  33. · Alex's posts · Alex Getman
    Excerpt
    🚨 The French at Mistral just dropped Large 4 Looks like Europe’s biggest hope in frontier AI is still alive. 1 trillion parameters, with 49B active at a time. In some benchmarks, it already beats GPT 6 Astra and competes with the strongest open weight models. The API is live now, with weights coming by the end of October. And yes, Mistral internally calls it Le Chonk. After the fat cat meme. https://www.youtube.com/watch?v=hD51W2txi1Y
  34. · Alex's posts · Alex Getman
    Excerpt
    🚨 Gemini 4 Argon scores 53 and ties for 3rd place with Fable 5.1 and GPT 6 Astra on the AA Intelligence Index
  35. · Alex's posts · Alex Getman
    Excerpt
    🚨 Sonnet 5.5 Max beats GPT-6 Astra Max on the Intelligence Index vs. cost curve A few other things stand out: 1. At medium, GPT-6 Sol is much cheaper than Sonnet 5.5 at a fairly similar intelligence level. 2. Around the Pareto line, GPT-6 Sol Max and Sonnet 5.5 High are basically neck and neck. 3. GPT-6 Astra Max lands roughly around Sonnet 5.5 XHigh. + Sonnet 5.5 Max is in a league of its own at the top.
  36. · Alex's posts · Alex Getman
    Excerpt
    🚨 Gemini 4.0 Pro specs leaked • Pricing: $2.25/$11.25 • 2M token context window • Beats GPT-6 Astra and Claude Fable 5.1 on almost every reported benchmark • While being significantly cheaper than both
  37. · Alex's posts · Alex Getman
    Excerpt
    🚨 Union Alpha just dropped in stealth on OpenCode It scores around 51% on Terminal Bench v4 at roughly $1.5 per task, putting it close to GPT 6 Astra high for much less. My guess is this is a new Chinese model, possibly Kimi. Kimi has also brought its paid plans back to the site after pausing new memberships. The timing lines up.
  38. · Alex's posts · Alex Getman
    Excerpt
    OpenAI has a new internal model that is way ahead of GPT 6 Astra in math On a curated set of open math problems, it reaches about 48% pass rate with more test time compute. GPT 6 Astra peaks at around 17%.
  39. · Alex's posts · Alex Getman
    Excerpt
    🚨 OpenAI’s new internal model solved the Navier-Stokes Millennium Prize Problem Around 10,000 agents worked on the solution for 88 hours, powered by a model significantly more capable than GPT 6 Astra. After the solution was found, GPT 6 Astra formalized and verified the proof in Lean.
  40. · Alex's posts · Alex Getman
    Excerpt
    🚨 Fable 5.2 could launch this month after Astra’s strong debut Anthropic is reportedly preparing its first fresh pretrain for Fable ahead of its IPO. People there are confident it can dethrone Astra. For context: Fable 5.1 and GPT 6 Astra currently share #1 on AA, with Fable slightly ahead in coding and Astra much stronger on 3D tasks.
  41. · Alex's posts · Alex Getman
    Excerpt
    🚨 GPT-6 Astra is officially #1 on the updated Artificial Analysis Intelligence Index, sharing the top spot with Claude Fable 5.1.
  42. · Alex's posts · Alex Getman
    Excerpt
    GPT 6 Astra set a new record in MazeBench, a 3D puzzle benchmark It explored 40% of all rooms and collected 14% of gems Fable 5.1 reached just 4.3% of rooms and collected 2% of gems
  43. · Alex's posts · Alex Getman
    Excerpt
    GPT-6 Astra Fast mode quietly went from 1.5x to 2x speed, with the same 2.5x usage multiplier. It’s still pretty expensive though. A task that would count as 1,000 usage units on Standard would count as 2,500 on Fast, so that’s still 150% more usage. *My previous post was deleted because I got the math wrong. I misled some of you, and I’m sorry about that.
  44. · Alex's posts · Alex Getman
    Excerpt
    Correction: I got the calculation in this post wrong. A 2.5× usage multiplier means 150% more usage, not 25%. Corrected calculation: https://alexgetman.com/403/gpt-6-astra-fast-mode-quietly-went-from-1-5x-to-2x-speed-with-the-same-2-5x-usag/ GPT-6 Astra Fast mode quietly went from 1.5x to 2x speed, with usage still at 2.5x So you’re getting double the speed for just 25% extra usage. At this point, there’s very little reason to turn Fast mode off.
  45. · Alex's posts · Alex Getman
    Excerpt
    GPT 6 Astra and Fable 5.1 got 20 attempts to pick up a block and place it into a bowl Astra: 95% success, 2.5 min, $0.94 per run Fable 5.1: 40% success, 6.8 min, $2.12 per run Astra also used 6× fewer output tokens, just 2.1k per run.
  46. · Alex's posts · Alex Getman
    Excerpt
    Claude is no longer the best model for frontend GPT-6 Astra just took the #1 spot, knocking Claude Fable 5.1 off the top The jump is massive: the previous GPT generation, GPT-5.6 Sol, was #13 A GPT model hasn’t led frontend since August 2025
  47. · Alex's posts · Alex Getman
    Excerpt
    GPT 6 Astra and Fable 5.1 got 20 attempts to pick up a block and place it into a bowl Astra: 95% success, 2.5 min, $0.94 per run Fable 5.1: 40% success, 6.8 min, $2.12 per run Astra also used 6× fewer output tokens, just 2.1k per run.
  48. · Alex's posts · Alex Getman
    Excerpt
    🚨 Plus users got access to GPT-6 Astra
  49. · Alex's posts · Alex Getman
    Excerpt
    🚨 GPT-6 Astra is out for Pro users
  50. · Alex's posts · Alex Getman
    Excerpt
    GPT-6 Astra looked like a generational leap on OpenAI’s benchmarks. Artificial Analysis puts it closer to old Fable 5 territory.
  51. · Alex's posts · Alex Getman
    Excerpt
    🚨 GPT-6 Astra is officially out https://openai.com/index/gpt-6-astra/
  52. · Alex's posts · Alex Getman
    Excerpt
    🚨 GPT-6 Astra is absolutely crushing everyone on the benchmarks
  53. · Alex's posts · Alex Getman
    Excerpt
    🚨 GPT-6 Astra has been added to the OpenAI API Launch in less than 24 hours
  54. · Alex's posts · Alex Getman
    Excerpt
    🚨 OpenAI is launching GPT 6 Astra next week It is a brand new model trained from scratch and the company’s largest since GPT 4.5. The latest internal build, codenamed mewfour, is already the release candidate.

    The forecast timing did not hold; retained as an early mention, not a successful prediction.

Alex's take

Signals 1

Specifications

GPT-6 Astra is a model from OpenAI, released on 3 Sep 2026. The context window is 1,050,000 tokens. The maximum output is 128,000 tokens. Pricing is $10.00 per million input tokens and $50.00 per million output tokens.

Context window
1,050,000
Max output
128,000
Input
text · image
Output
text
Input · USD / 1M tokens
$10.00
Output · USD / 1M tokens
$50.00
Cached input · USD / 1M tokens
$1.00

Questions

What is the context window of GPT-6 Astra?

The context window is 1,050,000 tokens. The maximum output is 128,000 tokens.

How much does GPT-6 Astra cost?

Pricing is $10.00 per million input tokens and $50.00 per million output tokens. Cached input is $1.00 per million.

When was GPT-6 Astra released?

On 3 Sep 2026.