AI Aulis

AI NEWS IN ENGLISH · 22 November 2026

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

21 stories · next issue tomorrow morning · How the paper is made →

← Issue 22 November 2026Paper

OpenAI: GPT-6.2 Astra writes more concisely than earlier models, so review your conciseness instructions

AI-generated contentWritten by AI

openai.com, the-decoder.com, searchenginejournal.com, cnet.com, developers.openai.com · 5 sources · Sources published: 19–21 November 2026 · Some source publication dates are unknown

OpenAI published ready-made prompts for GPT-6.2 Astra and Sol. They steer the models to make assumptions instead of asking, and to write new tests only when needed.

OpenAI has added ready-made prompts to its developer guide for the GPT-6.2 models. The prompts reflect how the company has seen Astra and Sol behave. The guide calls them a starting point and says they are worth trying if the model behaves the same way in your own application. The models were released on 17 November. Astra is designed for the most demanding tasks, and Sol for complex work at lower cost.

OpenAI says Astra writes more concisely and directly by default than earlier models, so ‘be concise’-style instructions are worth reviewing when switching models. Search Engine Journal notes that the GPT-6 Astra guide described that model as prone to detailed, heavily formatted answers.

When the user’s intent is unclear, Astra may propose a next step and wait for permission. The template instructs the model to make reasonable, reversible assumptions, tell the user about them and carry on working. The model should stop only if the task scope changes significantly, the work is concretely blocked or the user needs to make a decision. It should also stop before any irreversible or destructive action.

The coding template tells the model to use logic already in the repository and to avoid unnecessary helper functions and abstractions. It should write new tests only if they are requested, required by the repository or needed to catch a regression that existing tests would miss, the template says. The Decoder reported that GPT-6 Astra tested even small changes thoroughly.

CNET reports that some early testers have criticised the models. Ethan Mollick, an associate professor at the Wharton School, wrote on LinkedIn that Astra was harder to steer than competitors’ models and often seemed ‘lazy’. Thibault Sottiaux, head of Codex, said on X that influencers had received early versions and that the models have since been changed. Sottiaux suspects the influencers have not tested the models again.

Developers changing their prompts should also allow for an API change. Starting with GPT-6.2, the Responses API drops earlier reasoning if the conversation history is edited. Answer quality can then decline and the model may use more tokens. OpenAI recommends leaving the start of the conversation unchanged and adding new instructions at the end.

In the API, the models are called gpt-6.2-astra and gpt-6.2-sol. In ChatGPT Work and Codex, they are available to Plus, Pro, Enterprise and Edu users. Enterprise and Edu users get access once a workspace admin enables the models. Astra costs $8 per million input tokens and $40 per million output tokens for up to 272,000 tokens. Above that limit, Astra’s prices rise to $16 and $60, The Decoder reported. Sol costs $2 and $10.

For new integrations, the guide recommends the medium reasoning effort. It advises anyone using none or minimal to switch to low. The models do not support the temperature, top_p, logprobs and top_logprobs parameters. Chat Completions works, but tool calls require the Responses API.

Report an error in this story