GPT-6 Astra

Sep 2026
GPT-6 Astra
Compare in
on
$1.73per task?
$10 / $50per 1M in / outcache $1
59 tok/sfirst token in 64 s?
1.05Mcontext

Rankingat high effort?

Full ranking →
1Opus 5.561.42Fable 5.159.13Gemini 4 Argon57.14Fable 555.55Sonnet 5.555
···
7Muse Wonder Full53.68GPT-6 Astra53.19GPT-6.1 Sol52.9

Strong at

S
Coding#7 of 132Reasoning#8 of 334Science#2 of 312Knowledge#2 of 318Vision#2 of 124
A
Agents & work tasks#26 of 175Long context#41 of 273

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperClaude Opus 5.5medium+3%-23%$1.34Context: 1.05M → 1M+6OpenMiMo v2.6 Prothinking-7%-92%$0.13Context: 1.05M → 1M+4Lower effortSame modelmediumsame-11%$1.54CheaperGPT-6.1 Solmaxsame-58%$0.72Remove unsupported parameters+5CheaperMuse Spark 1.3max-8%-7%$1.60Context: 1.05M → 1M+6

Benchmarks

in Chat & Knowledge
AA Intelligence Index 4.3.250.9#8
AA Omniscience Index43.7#2
AA Omniscience · Accuracy61.1#5
AA Omniscience · Hallucination Rate44.8#74
Humanity's Last Exam53.1#8
GPQA Diamond94.9#2
AA-LCR (long context) 1.180#41
CritPt (physics)28.9#12
GDP.pdf · AA all-pass31#3
ARC-AGI-1 · public evaluation98.5#6
ARC-AGI-1 · semi-private98.5#1
ARC-AGI-2 · public evaluation97.9#2
ARC-AGI-2 · semi-private92.1#4
ARC-AGI-3 · semi-private54.8#1
FrontierMath Tier 4 (v2)97.6#2
AA Cyber Index successeslab33
HealthBench Professional length-adjustedlab70.3
HealthBench length-adjustedlab58.3
HealthBench Hard length-adjustedlab36.6
HealthBench Consensus length-adjustedlab95.5

Watch out 3

Monitoring can return HTTP 403›

Misalignment monitoring can return HTTP 403 before streaming or errors after output; match misalignment_policy_violation and handle stream errors. Source ↗

No none reasoning effort›

The none reasoning effort is unsupported on GPT-6 Astra and GPT-6.1 Sol; use low instead. Source ↗

Remove unsupported sampling parameters›

When reasoning effort is not none, remove temperature, top_p and top_logprobs, plus Chat Completions logprobs or Responses output logprobs inclusion. Source ↗

FAQ

How much does GPT-6 Astra cost?›

$10 per 1M input tokens and $50 per 1M output tokens; cached input $1 per 1M. A typical benchmark task costs about $1.73 at high effort.

Is GPT-6 Astra free to use?›

There is no free API; the cheapest paid price is $10 per 1M input tokens.

How good is GPT-6 Astra, and what is it best at?›

#8 of 334 in Chat & Knowledge, #7 of 132 in Coding, #9 of 133 in Agents & Tool Use, #26 of 195 in Writing & Language, #2 of 124 in Vision & Documents. Strongest at: Coding, Agents & work tasks, Reasoning, Science, Long context, Knowledge, Vision.

Is GPT-6 Astra better than Claude Opus 5?›

On Humanity's Last Exam, GPT-6 Astra scores 54.7 vs 54.9 — about the same. It is 71% cheaper per task.

How can I use GPT-6 Astra?›

Through the API of Amazon Bedrock, Azure, OpenAI.

What is the context window of GPT-6 Astra?›

1.05M tokens, with up to 128K output tokens.

When was GPT-6 Astra released? Is there a newer version?›

Released 2026-09-04. It is the current version in its line.

Is GPT-6 Astra open source?›

No, GPT-6 Astra is a closed model available through an API. The closest open-weights option here is MiMo v2.6 Pro.

Can GPT-6 Astra read images?›

It accepts text and images, PDF files as input.

What is GPT-6 Astra?›

GPT-6 Astra is a closed language model from OpenAI, released Sep 2026.

What are the best alternatives to GPT-6 Astra?›

Claude Opus 5.5 (cheaper), MiMo v2.6 Pro (open weights).