GPT-6 Astra

OpenAI's current model for demanding multistep work, including software engineering and tool-supported tasks. These notes describe model behavior, not Codex behavior.

• State the target outcome, acceptance criteria, constraints, and relevant context.
• Specify required output structure and genuine invariants.
• Define when to ask for missing evidence, when a reasonable assumption is allowed, and when the task is complete.
• If requirements change during a long task, state the new requirement directly and name any invariant that must remain unchanged.
• Define the validation scope explicitly for small tasks; Astra may test more broadly than the task requires.

• Prescribing every procedural step when only the result is constrained.
• Leaving success conditions implicit for long-running work.
• Assuming the model will infer a preferred response style when formatting or verbosity matters.

Model behavior is nondeterministic. GPT-6 Astra does not support the none reasoning effort, and supported effort choices depend on the calling API. Evaluate guidance against representative work and update it when the official model guidance changes.