OpenAI
GPT-6 Astra
OpenAI frontier flagship with Critical-tier autonomous cybersecurity, extreme token efficiency, and 72.6% on OSWorld 2.0.
Intelligenza
86.2 #39
AAQI 61.0
Velocità Uscita
82 t/s
TTFT 143 ms · ITL 12.2 ms/tok
Prezzo combinato (3:1)
$20 /1M
Input $10/M · Output $50/M
Contesto
1.1M
≈ 2,750 pagine A4 · output max 131,072
Quality Domain Radar
6-Axis Frontier Evaluation vs Category Average & Flagship
GPT-6 Astra
Claude Mythos Preview (#1)
Average (68)
Gli assi radar sono approssimati dai punteggi di categoria dove mancano misurazioni dirette — vedi Metodologia.
Punti di forza e compromessi
Dove vince
- 72.6% OSWorld 2.0 — best-in-class computer use
- Flagship agentic + reasoning scores across the board
- 131K output tokens — rewrites entire files in one pass
- Strong tool-use reliability on terminal workflows
A cosa fare attenzione
- $10/$50 premium pricing tier
- Higher effort levels multiply latency and token spend
- Closed weights — no self-hosting
Casi d'uso killer
OSWorld computer-use agents
Desktop-level control — clicks, typing and app workflows at a benchmark-leading 72.6%.
Terminal-Bench autonomous ops
Long shell sessions: builds, tests, fixes, deploys.
Large-repo refactoring
131K output window handles whole-file and multi-file rewrites.
Quickstart per sviluppatori e API
Imposta prima
OPENAI_API_KEY — ottieni una chiave APIcurl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "content-type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": "Refactor this function for O(n) time and explain the change."
}'from openai import OpenAI
client = OpenAI() # reads OPENAI_API_KEY
resp = client.responses.create(
model="gpt-6-astra",
input="Refactor this function for O(n) time and explain the change.",
)
print(resp.output_text)const res = await fetch("https://api.openai.com/v1/responses", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.OPENAI_API_KEY}`,
"content-type": "application/json",
},
body: JSON.stringify({ model: "gpt-6-astra", input: "Refactor this function for O(n) time and explain the change." }),
});
console.log((await res.json()).output_text);Risorse e link ufficiali
Deployment API
Confronta host e latenza
Throughput, TTFT e prezzi cache su Groq, Cerebras, Together, DeepInfra, Fireworks, Bedrock e Azure.
GroqCerebrasTogether AIDeepInfraFireworks AIAWS BedrockAzure AI
Punteggi per Fonte
Punteggi per Categoria
Risultati dei Benchmark
Specifiche live e architettura
Sotto il cofano
GPT-6 Astra — undisclosed transformer
- Reasoning effort levels: low / medium / high / xhigh — one weight set, four budgets
- Critical-tier autonomous cybersecurity designation
- 1.1M context with a 131K max output window
- Extreme token efficiency on long agent runs
Data di Rilascio
3 set 2026
Architettura
GPT-6 Astra — undisclosed transformer
Parametri
Not disclosed
Licenza
Proprietary
Finestra di Contesto
1,100,000 tokens
Output max
131,072 tokens
Modalità
text + image
Prezzi cache
lettura $1 · scrittura $12.5 /1M
Ragionamento
Livelli di sforzo
Prezzo Ingresso
$10/M
Prezzo Uscita
$50/M
Velocità Uscita
82 t/s
Tempo al Primo Token
143 ms
Dati aggiornati
4 set 2026
Stima bolletta mensile
Spesa mensile stimata
$0.00
Calcolo risparmio vs GPT-4o…
Cita e incorpora
Badge Markdown GitHub
Loading badge markdown...BibTeX
@misc{llmpodium2026gpt6astra,
title = {Benchmark Analysis and Intelligence Evaluation of GPT-6 Astra},
author = {LLMPodium Research},
year = {2026},
url = {https://llmpodium.com/models/gpt-6-astra}
}Incorpora il badge del punteggio
Aggiungi un badge Podium Score live al tuo README su GitHub o Hugging Face — rimanda alla classifica live.
Anteprima
Markdown
Loading badge snippet...HTML
Loading badge snippet...Altri modelli di OpenAI
Modelli alternativi
Migliori alternative a GPT-6 Astra in base a capacità, prezzi e benchmark.