Text model
nemotron-3-ultra-nvfp4 pricing
Fireworks AI charges $0.60 per million input tokens and $2.40 per million output tokens for nemotron-3-ultra-nvfp4, and $0.12 for cached input. Below: every rate that applies, what common workloads cost, and cheaper options.
- Input
- $0.60/ 1M
- Output
- $2.40/ 1M
- Cached input
- $0.12/ 1M
- Context
- 262.1Ktokens
Retires 2026-11-07. npx @veqen/sdk scan finds every place your code uses it.
All rates
| Tier | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
| Standard | $0.60 | $2.40 | $0.12 | — |
| Priority | $0.75 | $3.00 | $0.15 | — |
USD per million tokens. Long-context tiers apply when the whole prompt, cached tokens included, passes the threshold.
Work it out
What common workloads cost
| Workload | Per call | Cost per call | Calls a month | Per month |
|---|---|---|---|---|
| Chat reply | 1,500 input and 400 output tokens | $0.00186 | 100,000 | $186.00 |
| RAG answer, cached context | 2,000 fresh + 4,000 cached input, 500 output | $0.00288 | 100,000 | $288.00 |
| Agent step | 4,000 fresh + 8,000 cached input, 800 output | $0.00528 | 250,000 | $1,320.00 |
| Batch summary | 3,000 input and 300 output tokens via the Batch API | — | 1,000,000 | — |
Planning a mix of models? Simulate a workload with nemotron-3-ultra-nvfp4, alongside anything else you run.
Cheaper options
Current models that cost at least 15% less for a typical chat reply. Cheaper isn't always equivalent: test quality on your own prompts, then use veqen diff to see the impact at your volume.
| Model | Chat reply | Saving |
|---|---|---|
| fireworks/accounts/fireworks/models/deepseek-v3p1 | $0.00151 | 19% |
| fireworks/accounts/fireworks/models/deepseek-v3p1-terminus | $0.00151 | 19% |
| deepinfra/moonshotai/Kimi-K2.5 | $0.00158 | 15% |
| deepinfra/moonshotai/Kimi-K2-Instruct | $0.00155 | 17% |
Price history
No price changes since Veqen started tracking this model.
Compare
Last verified with price book 2026-10-22. Sources: Fireworks AI pricing and LiteLLM's MIT-licensed price file. Track every call to this model, per customer or project: install the SDK.