GPT-6 Astra

Sep 2026
GPT-6 Astra
Compare in
on
$3.26per task?
$10 / $50per 1M in / outcache $1
69 tok/sfirst token in 4 min?
1.05Mcontext

Rankingat max effort?

Full ranking →
1Opus 5.561.42Fable 5.159.13Gemini 4 Argon57.14Fable 555.55Sonnet 5.5556Opus 554.97GPT-6 Astra54.78Muse Wonder Full53.6

Strong at

S
Coding#3 of 132Reasoning#7 of 334Science#1 of 312Knowledge#3 of 318Vision#2 of 124
A
Agents & work tasks#25 of 175Long context#36 of 273

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperGemini 4 Argonhigh+4%-39%*$1.99Context: 1.05M → 1M+3OpenMiMo v2.6 Prothinking-10%-96%$0.13Context: 1.05M → 1M+4Lower effortSame modelxhighsame-29%$2.31CheaperClaude Opus 5.5highsame-44%$1.82Context: 1.05M → 1M+6CheaperClaude Fable 5.1mediumsame-8%$2.98Context: 1.05M → 1M+6

Benchmarks

in Chat & Knowledge
AA Intelligence Index 4.3.252.7#5
AA Omniscience Index43.4#3
AA Omniscience · Accuracy62.6#4
AA Omniscience · Hallucination Rate51.3#92
Humanity's Last Exam54.7#7
GPQA Diamond96.1#1
AA-LCR (long context) 1.180.7#36
CritPt (physics)31.7#3
GDP.pdf · AA all-pass31#3
MLCR-AA35#13
ARC-AGI-1 · public evaluation97.8#12
ARC-AGI-1 · semi-private97.5#7
ARC-AGI-2 · public evaluation97.5#3
ARC-AGI-2 · semi-private95#1
ARC-AGI-3 · semi-private62.7#1
Arena Text · overall1,477#23
Arena Text · overall · full cohort1,442.7#46
Arena Text · Chinese1,493#46
Arena Text · English1,447.9#52
Arena Text · French1,444.4#64
Arena Text · German1,457.1#31
Arena Text · instruction following1,454.7#34
Arena Text · Japanese1,392.3#51
Arena Text · Korean1,429.6#24
Arena Text · non-English1,432.5#43
Arena Text · Polish1,419.5#57
Arena Text · Russian1,437#44
Arena Text · Spanish1,422.6#68
GPQA Diamond · Epoch95.8#2
OTIS Mock AIME 2024 2025100#1
FrontierMath Tiers 1–3 (v2)93.7#1
FrontierMath Tier 4 (v2)97.6#2
SimpleQA Verified v175.6#3
Chess Puzzles74#1
Mystery Game Puzzles84#1
FrontierMath Erdos2.9#2
MathArena · ArXivMath · April 202690.8#1
MathArena · ArXivMath · August 202688.6#3
MathArena · ArXivMath · June 202694.4#1
MathArena · ArXivMath · May 202695#1
MathArena · BrokenArXiv · April 202697.5#1
MathArena · BrokenArXiv · August 202681.9#2
MathArena · BrokenArXiv · June 202699.1#3
MathArena · BrokenArXiv · May 202691#2
MathArena · ArXivLean · June 202665.2#1
MathArena · ArXivLean · September 202637.7#3
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 $3.26 at max 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?›

#7 of 334 in Chat & Knowledge, #3 of 132 in Coding, #6 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 44% 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?›

Gemini 4 Argon (cheaper), MiMo v2.6 Pro (open weights).