---
title: GPT-6.2 Astra
description: OpenAI’s flagship GPT-6.2 model, built for the hardest projects —
  real-world engineering, professional work, and scientific research, with
  stronger visual reasoning for websites and slideshows.
canonical: https://productized.tech/models/gpt-6-2-astra
publishedAt: 2026-11-17
lang: en
---

# GPT-6.2 Astra

Developer: OpenAI

Reasoning used: provider-default

## Specifications

- Released: 17 Nov 2026

- Developer: OpenAI

- Context window: 1.05M tokens

- Modalities: text, image

- Pricing / 1M: $8 in · $40 out

- Reasoning depth: low, medium, high, xhigh

## Creative writing

### A Second Opinion for Everyone

A billion people with advanced AI could mean a billion people able to get a second opinion. It could also mean a billion people getting the same second opinion, delivered in a voice uncannily like their own. I think the difference between those futures matters more than how clever the next model is.

Consider a tenant whose heat goes out in January. An assistant can help her read the lease, document the cold, and draft a letter that once would have cost an afternoon’s wages in legal fees. That is a real gain in freedom. But her landlord can buy an assistant too. If it can generate a plausible reason to reject every complaint in seconds, while each appeal costs the tenant a morning off work, the smarter machines have made her worse off. The scarce resource was never a well-written letter. It was somebody with the power and obligation to act on it.

I would start there. Whenever an institution uses AI to make a decision that can cost you your home, livelihood, or liberty, you should be able to see the evidence, correct the record, and reach a named person authorized to reverse it. Let institutions automate paperwork. Don’t let them automate away the moment when someone has to answer for a decision.

Thriving takes more than protection against a bad decision. I’d put public money into places where people already bring unfinished questions: libraries, clinics, night schools. Give a librarian a budget to try different assistants and paid time to sit beside a patron. Together they can check whether the impressive quotation from the housing code is real and, more importantly, find the inspector who can get the heat turned on.

My test for Mass Intelligence is deliberately ordinary. On the coldest night of the year, does the tenant get a more eloquent refusal, or does she get heat? We should build for the second outcome and refuse to count the first as progress.

## Landing page

[Open landing page](https://artifacts.productized.tech/models/gpt-6-2-astra/landing/index.html)

## Presentation

[Download presentation](https://artifacts.productized.tech/models/gpt-6-2-astra/presentation/deck.pptx)

![ONE PIECE OF ADVICE](https://artifacts.productized.tech/models/gpt-6-2-astra/presentation/slides/01.png)

![IN PRACTICE](https://artifacts.productized.tech/models/gpt-6-2-astra/presentation/slides/02.png)

![WHY THIS MATTERS](https://artifacts.productized.tech/models/gpt-6-2-astra/presentation/slides/03.png)
