| Nemotron 3.5 Lightning | GPT-5.4 Mini | |
|---|---|---|
| Input, per 1M tokens | $0.08 | $0.75 |
| Output, per 1M tokens | $0.20 | $4.50 |
| Context window | 1M | 400K |
| Classification or routing (600 tokens in, 10 out), per 1,000 | $0.05 | $0.495 |
| Chat reply (2,000 in, 400 out), per 1,000 | $0.24 | $3.30 |
| Retrieval-augmented answer (8,000 in, 500 out), per 1,000 | $0.74 | $8.25 |
| Confidence signal | yes | no |
| Tool calling | yes | yes |
| BANKING77, correct of 400 | 255 | not measured |
Which one should carry your traffic?
If Nemotron 3.5 Lightning answers your requests as well as GPT-5.4 Mini, moving that traffic saves 93% on chat-shaped requests. Averages cannot tell you whether it does for your requests. Nemotron 3.5 Lightning exposes a confidence signal, so Gimbix can let it answer first and hand unsure cases to GPT-5.4 Mini.
Nemotron 3.5 Lightning profile · GPT-5.4 Mini profile · Measure it on your traffic