GPT-6 Astra
Research Engineer at OpenAI | Complex reasoning, coding and computer use | 1M-token context
The basics
Checked Oct 20- Status
- Model ID
- Price
- Context
Updates
Strengths and weaknesses
Strengths
Weaknesses
- !
Role
Flagship model at OpenAI
Sep 2026 to now · 2 months- Sep 3, 2026: Frontier model
Apr 2026 to Sep 2026 · 5 months
Mar 2026 to Apr 2026 · 1 month
Works with
From recorded changes
Agent · 1 shared change
Agent · 1 shared change
Agent · 1 shared change
App · 1 shared change
API · 1 shared change
Similar models
$4 in, $20 out · Compare
$2 in, $10 out · Compare
$2 in, $10 out
Cost vs peers
Most expensive of 6 similar models by output price.
Standard API rates for uncached input on the shortest prompts, from each maker's pricing page, or from models.dev for models not researched here yet. Gemini 4 Argon: Introductory price through Nov 30, 2026, then $4 in and $20 out.
Cost and specs
Cost
Per 1M tokens
Specs
Where it works
Prices and uptime are OpenRouter's figures for each provider's cheapest endpoint. Uptime covers the last 24 hours.
Benchmarks
Higher is better. Green is GPT-6 Astra. Dark marks are similar models.
Independent
What people say
Reviews (1)
Lior GrossmanShipping software since 1993, now building with AIGPT-6 Astra is my second opinion for reviews, math and experiments. Before I commit to an architecture, an experiment or a strategy, I ask Astra to find what's wrong with it, and it usually finds something. I made an Astra review a required step before my coding agents commit. It once caught a bug that erased a user's answers without asking, before the add-on reached users. It's like a person with an IQ of 200 and no social skills. It's brilliant at adversarial review, statistics and math. It sucks at UI and copywriting, so I never consult it on copy. It also nitpicks, overengineers and causes scope creep, so I tell my agents to take only its important fixes.
Oct 31, 2026
What builders said
Short excerpts from public posts, each linked to the original.
"Astra is almost unusable, if i ask one prompt it burns 10-20% of my weekly, but my project is very complicated; even using sol 6.1 on high i can easily use the whole 100/m plan weekly in a day."
"It found out over 60 bugs and also very token efficient and after that I've done all reviews with Astra only and I'm happy with results :)"
Incidents
Incidents per week, last 13 weeks. Most in one week: 1.
Writing habits
Creative writing · EQ-Bench, among 51 models tracked here
Slop score
8.4
Overused AI words and phrases. Less slop than 96% of tracked models.
Repetition
3.5
Repeated words and ideas. Less repetition than 64% of tracked models.
Story length
6,191
Average characters per story. Longer than 52% of tracked models.