GPT-5.4 vs GPT-5 mini
Use GPT-5 mini by default. Upgrade to GPT-5.4 only for tasks where eval data proves mini fails.
GPT-5.4 is 10× more expensive than GPT-5 mini ($2.50 vs $0.25/M input). GPT-5 mini covers most production use cases — classification, extraction, Q&A, standard coding. Use GPT-5.4 for complex reasoning, strict code review, and tasks where mini's 8-12% lower benchmark scores actually translate to failures in your workload.
Side-by-side pricing
| GPT-5.4 | GPT-5 mini | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Input price/1M | $2.50 | $0.25 |
| Output price/1M | $15.00 | $2.00 |
| Cache read/1M | $0.25 | $0.025 |
| Context window | 1M | 400K |
| Vision | ✓ | ✓ |
| Tool use | ✓ | ✓ |
| Prompt caching | ✓ | ✓ |
| Speed tier | Standard | Fast |
Prices verified 2026-08-20. See changelog for history.
Winner by task
Calculate costs for each model
Frequently Asked Questions
Is GPT-5.4 worth 10× the cost?
Rarely. Benchmark improvements don't translate linearly to task success improvements. For most tasks, GPT-5 mini gets you 90% of the quality at 10% of the cost. Run your specific eval before deciding — don't assume you need GPT-5.4.
Prices verified 2026-08-20. See methodology for calculation details.