GPT-5.4 Nano
| Maker | OpenAI |
|---|---|
| Released | 2026-03-17 |
| Context | 400K |
| Max output | 128K |
| Open weights | Closed |
| Modalities | text, image → text |
|---|---|
| Input / 1M | $0.20 |
| Output / 1M | $1.25 |
| Cached input / 1M | $0.02 |
| Review call | $0.0083 |
| Reasoning efforts | low, medium, high, xhigh |
|---|---|
| Hosts | OpenAI, OpenRouter |
| Model id | gpt-5.4-nano |
CodePanel's benchmark has not run GPT-5.4 Nano as a pull-request reviewer yet, so this page has no recall, false-alarm or measured-cost figures for it. The methodology explains how ranked models are scored.
No GPT Nano model is ranked yet. Of OpenAI's ranked models, GPT-OSS 120B (Workers AI) is the closest in list price; it caught 40% [29%–52%] of the known bugs at rank 8 of 17, with a measured review cost of $0.0075. At list price, a typical review call (26.2K tokens in and 2.4K out) costs $0.0083 on GPT-5.4 Nano and $0.011 on GPT-OSS 120B (Workers AI), so this model's call is 25% less than that one's.
| Host | Model id | Input / 1M | Output / 1M |
|---|---|---|---|
| OpenAI | gpt-5.4-nano | $0.20 | $1.25 |
Source: OpenAI pricing, Oct 16, 2026
| Benchmark | Score | Range | Setting | Source | Date |
|---|---|---|---|---|---|
| Epoch Capabilities Index | 146 ECI | 144 to 149 | – | Epoch AI CC BY 4.0 | Mar 17, 2026 |
Same as on GPT-5.6 Luna.
Updated Oct 23, 2026