Blog

New OpenAI Models in AI Game Developer: GPT-6.2 Astra and GPT-6.2 Sol

OpenAI GPT-6.2 Astra and OpenAI GPT-6.2 Sol are now in AI Game Developer. What is new, why Flex matters, and how to pick the right model.

New OpenAI Models in AI Game Developer: GPT-6.2 Astra and GPT-6.2 Sol

Two new OpenAI models are now in AI Game Developer. OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, and both are in the model list of the app today. This post explains what each new model is good at, how it differs from the models you already use, how it counts against your plan limit, and how to choose between them.

If you are new to AI Game Developer, the quickstart at ai-game.dev/docs/quickstart connects your Unity, Unreal Engine or Godot project in a few minutes.

The New OpenAI Models at a Glance

  • GPT-6.2 Astra (OpenAI): OpenAI's new frontier model, for the hardest reasoning, coding and design work. It joins the list next to GPT-6 Astra.
  • GPT-6.2 Sol (OpenAI): the new everyday Sol model, built for fast multi-step agent work. It joins the list next to GPT-6.1 Sol.

Both accept text and images, so you can paste a screenshot of your scene or your UI into the chat and talk about it.

GPT-6.2 Astra: OpenAI's Frontier Model

Astra is the model to reach for when a task is genuinely hard: a combat system that touches many scripts at once, a tricky physics bug, a save system that has to survive game updates, or a large refactor. GPT-6.2 Astra is a clear step up from GPT-6 Astra. OpenAI reports that it scores 6.2 points higher on the GDPval v2 benchmark of professional work while using 22 percent fewer output tokens, and 9 points higher on Terminal-Bench-Science, a test of long tasks in a real terminal.

One result stands out for game developers. On AA-3D, a benchmark where human judges compare two blinded interactive 3D experiences built from the same prompt, GPT-6.2 Astra reached an Elo score of 1318, ahead of Claude Fable 5.2 at 1271. OpenAI also says Astra is stronger at visual reasoning, so it is a good partner for level layouts, procedural geometry, shaders and anything that has to look right, not only compile.

Astra has a large context window of 922,000 input tokens and can write up to 128,000 tokens in one answer, so it can keep a big part of your project in mind at once.

GPT-6.2 Sol: The Everyday Workhorse

Sol is OpenAI's efficient model for everyday and multi-step agent work: writing gameplay scripts, wiring up UI, fixing compile errors, creating prefabs and scenes through the tools of AI Game Developer, and iterating quickly. GPT-6.2 Sol scores 5.3 points higher than GPT-6.1 Sol on GDPval v2, and on the FrontierCode 2.0 coding benchmark it lands within half a point of Astra, at a much lower cost per token than Astra.

Sol has the same context window as Astra: 922,000 input tokens and up to 128,000 output tokens.

Why GPT-6.1 Sol and GPT-6 Astra Stay in the List

Usually a new version simply replaces the old one in AI Game Developer. This time we keep both, and here is why.

Our OpenAI models run in the Flex class by default. Flex is OpenAI's cheaper and somewhat slower service class, and it uses about half as much of your plan limit per token as the Standard class. Many of you use Flex every day because your plan lasts longer with it. OpenAI no longer offers Flex for new models, and the GPT-6.2 models are Standard only: we tested it, and the API refuses a Flex request for both.

So you get to choose:

  • GPT-6.1 Sol and GPT-6 Astra with Flex: they use the least of your plan limit, and you can still switch them to Standard in the model picker when you want faster answers.
  • GPT-6.2 Sol and GPT-6.2 Astra: the newest and smartest versions, always in the Standard class, so the picker shows no Flex/Standard switch for them. GPT-6.2 Sol draws about twice as much of your plan limit per token as GPT-6.1 Sol on Flex. GPT-6.2 Astra draws about 1.6 times as much as GPT-6 Astra on Flex for new input and output, and a little less for reused cached context. Compared with GPT-6 Astra on Standard, it draws about 20 percent less.

Nothing changes for your OpenAI models unless you pick a new one. If GPT-6.1 Sol or GPT-6 Astra is your model today, it stays your model, at the same usage as before.

Which Model Should You Pick?

  • Big feature, hard bug, or 3D and visual work: GPT-6.2 Astra.
  • Long everyday sessions where your plan limit matters most: GPT-6.1 Sol with Flex.
  • Everyday work where you want the newest model and faster answers: GPT-6.2 Sol.
  • Quick, light tasks on any plan: GPT-6 Luna.

A good rhythm for bigger features is to let a strong model plan the work, then let a faster model carry out the steps. Plan mode, described at ai-game.dev/blog/plan-mode-ai-agent-game-development, is made for exactly that.

Who Can Use the New Models

GPT-6.2 Astra and GPT-6.2 Sol are part of the Studio plan and higher, like the other OpenAI models in the app except GPT-6 Luna. Plans are compared at ai-game.dev/pricing.

Every model uses your plan limit first, shown as a percentage of your allowance. When it runs out, work can continue from your extra balance if you turned on Extended usage on your account page. The live list of models available in your region is at ai-game.dev/model-availability.

Get Started

  1. Download or update the AI Game Developer App from ai-game.dev/download and sign in.
  2. Open the model picker in a chat and choose GPT-6.2 Astra or GPT-6.2 Sol.
  3. Give it a real task from your game, and compare the result with the model you used before.

The header image is from OpenAI's announcement of GPT-6.2 Astra and Sol at https://openai.com/index/introducing-gpt-6-2-astra-sol/ where you can read the full details and benchmark results.

Thank you for building games with AI Game Developer. Try the new models on your project, and tell us in our Discord community which one works best for you.

More from the blog

All posts