GPT-5.4 mini and GPT-5.4 nano
2026-03-17 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI launches GPT-5.4 mini and GPT-5.4 nano, its most capable small models: mini targets fast coding and subagents, nano targets classification and extraction at very low cost.
What changes
- GPT-5.4 mini runs more than 2 times faster than GPT-5 mini and approaches GPT-5.4 on SWE-Bench Pro and OSWorld-Verified.
- Computer use, tool search and compaction are supported by mini; nano handles compaction only.
- In Codex, mini uses 30% of the GPT-5.4 quota and serves as a delegated subagent.
- Nano is API-only; mini is also available in ChatGPT (Free and Go through Thinking, limit fallback for the others).
### Specifications - Context: 400,000 tokens, maximum output 128,000 tokens, text and image input. - Knowledge cutoff: August 31, 2025. Effort: none (default) to xhigh. - Availability: API, Codex and ChatGPT for mini; API only for nano. - Identifiers: gpt-5.4-mini, gpt-5.4-nano. ### Pricing | Model | Input | Cached input | Output | | --- | --- | --- | --- | | gpt-5.4-mini | $0.75 | $0.075 | $4.50 | | gpt-5.4-nano | $0.20 | $0.02 | $1.25 | Price per million tokens (USD). ### Benchmarks | Evaluation (xhigh) | GPT-5.4 | mini | nano | GPT-5 mini | | --- | --- | --- | --- | --- | | SWE-Bench Pro (public) | 57.7% | 54.4% | 52.4% | 45.7% | | Terminal-Bench 2.0 | 75.1% | 60.0% | 46.3% | 38.2% | | OSWorld-Verified | 75.0% | 72.1% | 39.0% | 42.0% | | GPQA Diamond | 93.0% | 88.0% | 82.8% | 81.6% | | Toolathlon | 54.6% | 42.9% | 35.5% | 26.9% | | MMMU Pro | 81.2% | 76.6% | 66.1% | 67.5% | Scores published by the vendor. ### Key takeaways The "large model plans, small models execute" pattern becomes an official Codex usage. Mini delivers most of GPT-5.4 for about 30% of the input price.
Links
Tags : LLM, OpenAI, GPT-5, Mini, Nano, Sous-agents