GPT-5.1 Prompting Guide (cookbook.openai.com)

OpenAI published a GPT-5.1 prompting guide describing the new flagship’s design and practical prompt patterns for production agents. GPT-5.1 emphasizes a balance of intelligence and speed, adds a low-latency “non‑reasoning” mode, and dynamically calibrates token usage to prompt difficulty (fewer tokens for easy inputs, more for hard ones). It’s also more steerable for personality, tone, and output formatting, and a companion GPT-5.1‑codex variant is available for coding-focused workflows. The guide’s concrete recommendations include nudges for persistence (avoid overly terse answers that sacrifice completeness), explicit verbosity/output-format controls, and a migration note for coding agents to move apply_patch to a new named-tool API. Technically important implications: the model supports a dedicated verbosity parameter and strict final-answer formatting and code-output budgets (rules for when to include snippets, file-level summaries, and limits on before/after blocks). The guide prescribes user‑update/preamble conventions for agent rollouts (short 1–2 sentence updates every few tool calls or steps, explicit plans before first tool call, and recaps with status), an “immediacy” rule to send commentary before analysis, and a solution‑persistence policy that biases the model to complete end‑to‑end tasks rather than stopping early. Together, these patterns are aimed at improving reliability, latency, and human‑agent coordination in production deployments.