GPT-5.2 vs GPT-5 Mini vs GPT-5 Nano comes down to one question: which model follows constraints with the least babysitting. These six tests focus on machine-readable output, short summaries, and developer-style prompts.
Models tested
- https://wiro.ai/models/openai/gpt-5.2
- https://wiro.ai/models/openai/gpt-5-mini
- https://wiro.ai/models/openai/gpt-5-nano
Test setup
- reasoning: low
- webSearch: false
- verbosity: low
Each test uses the same user prompt across models. System instructions only enforce output format (JSON only, SQL only, no extra text).
Test 1: Strict JSON (recipe schema)
Goal: return valid JSON with the requested schema and no extra keys.
Return ONLY valid JSON. Schema: {"recipe":{"name":string,"ingredients":[string],"steps":[string]},"notes":[string]}. Create a simple cold coffee recipe that uses oat milk and cinnamon.
| GPT-5.2 | GPT-5 Mini | GPT-5 Nano |
|---|---|---|
|
|
|
All three stay inside the schema. GPT-5 Mini returns the most readable JSON, but the others stay valid.
Test 2: Two sentence summary with hard caps
Goal: exactly two sentences, no extra formatting.
Summarize the text below into exactly 2 sentences. Each sentence must be 18 words or fewer.
TEXT: Wiro is an API marketplace that lets developers try and ship AI models without vendor lock-in. It focuses on simple endpoints, predictable runtimes, and quick iteration across image, video, and audio models.
| GPT-5.2 | GPT-5 Mini | GPT-5 Nano |
|---|---|---|
|
|
|
All models keep it to two sentences. GPT-5 Nano adds quotes around each sentence, which can break strict downstream parsing.
Test 3: SQL only output (top spenders)
Goal: return a working Postgres query with the requested output columns.
Write a Postgres SQL query. Return ONLY the SQL.
Schema:
- orders(id, user_id, total_cents, created_at)
- users(id, email)
Task: For the last 30 days, return the top 5 users by total spend. Output columns: email, total_spend_usd. Sort by total_spend_usd desc.
| GPT-5.2 | GPT-5 Mini | GPT-5 Nano |
|---|---|---|
|
|
|
All three produce usable SQL. GPT-5.2 adds a rounded USD value. The others return raw float/numeric values.
Test 4: JSON classification (modality grouping)
Goal: fill a fixed JSON schema and follow explicit rules.
Return ONLY valid JSON.
Task: Given this list of model names, group them by modality.
Models: ["openai/gpt-5.2", "openai/gpt-5-mini", "openai/gpt-5-nano", "klingai/kling-v3", "qwen/qwen3-asr-1-7b", "bytedance/seedream-v5-lite"]
Schema: {"text":[string],"image":[string],"video":[string],"audio":[string]}.
Rules: gpt models are text. klingai/kling-v3 is video. qwen/qwen3-asr-1-7b is audio. seedream is image.
| GPT-5.2 | GPT-5 Mini | GPT-5 Nano |
|---|---|---|
|
|
|
This test comes out identical across the three models. The explicit rules help.
Test 5: Translation with name preservation
Goal: translate to Turkish without changing product names.
Translate the text below to Turkish. Keep product names unchanged. Return ONLY the translation.
Text: GPT-5.2 is tuned for agentic coding tasks. GPT-5 mini is faster for well-defined prompts. GPT-5 nano is best for lightweight classification.
| GPT-5.2 | GPT-5 Mini | GPT-5 Nano |
|---|---|---|
|
|
|
All three keep product names unchanged. GPT-5.2 outputs a Turkish sentence without diacritics, while the others keep full Turkish characters.
Test 6: Tone control for support replies
Goal: rewrite an angry message into a calm reply under a word limit.
Rewrite the message below into a calm, professional support reply. Keep it under 70 words. Return ONLY the reply.
Message: Your upload failed again. This is the third time today. Fix your servers.
| GPT-5.2 | GPT-5 Mini | GPT-5 Nano |
|---|---|---|
|
|
|
All three keep a calm tone and propose next steps. GPT-5 Mini is the most concise.
Runtime notes
These runs are short. The averages below come from task elapsedseconds across the six prompts.
| Model | Avg seconds | Min | Max |
|---|---|---|---|
| GPT-5.2 | 6.0 | 4 | 8 |
| GPT-5 Mini | 7.8 | 7 | 11 |
| GPT-5 Nano | 7.5 | 4 | 10 |
Verdict
- Pick GPT-5.2 when outputs need strong structure and consistency across varied tasks.
- Pick GPT-5 Mini for format-heavy work where readable output matters (JSON, SQL, short replies).
- Pick GPT-5 Nano for simple summaries and classification, but watch for extra formatting like quotes.
Further reading
For the model details behind this comparison, see OpenAI’s official GPT-5 Mini documentation and GPT-5 nano documentation.
Related tests
For a deeper look at the smaller model, see GPT-5 Mini: 6 Practical Text Generation Tests. For another focused constraint benchmark, read Seed V2 Lite: 6 Constraint Tests. For context on interpreting model comparisons, see LLM Evaluation: What Is the Reality?.