OpenAI — JSON Mode, Function Calling, Reasoning

~14 min · providers, openai

Level 0Apprentice

0 XP0/100 lessons0/14 achievements

0/120 XP to next level120 XP to go0% complete

What OpenAI does well

Strict JSON Schema mode (response_format with json_schema, strict: true).
Mature function calling with parallel calls.
o-series and GPT-5.5 reasoning with effort levels.
Multimodal across text, vision, audio.
Realtime API for low-latency voice.

Quirks to know

Three roles: system, developer, user. Developer is for app-level instructions that should outrank user but be invisible to users.
Reasoning tokens are billed but not returned to the developer (you see the answer, not the trace).
Strict JSON Schema mode supports a subset of JSON Schema — not all keywords are accepted.
Function-call style: arguments are returned as a JSON string you parse, not a parsed object.

Style preferences

OpenAI models respond well to terse, imperative system prompts and well-described function definitions. Heavy XML-tag scaffolding can interfere; keep it lighter than on Claude.

Code

OpenAI — strict structured output + reasoning·python

import openai

client = openai.OpenAI()
resp = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "system", "content": "You are a contract analyst."},
              {"role": "user", "content": question}],
    response_format={
        "type": "json_schema",
        "json_schema": {"name": "findings", "strict": True, "schema": SCHEMA}
    },
    reasoning={"effort": "high"},
    max_completion_tokens=2048,
)

External links

Exercise

Migrate one prompt from a non-strict 'please return JSON' style to OpenAI strict JSON Schema mode. Note the prompt simplifications you can make once the schema is enforced.

Progress

Progress is local-only — sign in to sync across devices.

← PreviousClaude — System Prompts, Long Context, XML Tags Next →Gemini — Multimodal, Long Context, Reasoning

Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)

No comments yet — be the first.