Best cost/speed trade-off
DeepInfra
deepinfra/fp4$1.0667 / 1M tokens
14.97 s estimated response
NVIDIA: Nemotron 3 Ultra provider comparison
nvidia/nemotron-3-ultra-550b-a55b
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...
- Endpoints:
- 3
- Context:
- 1M
- Input:
- text
- Output:
- text
At a glance
Comparison winners
Based on 1,000 input and 500 output tokens using the latest published median speed data.
Visual comparison
Price and performance charts
Compare published endpoint pricing and estimated response time visually.
USD per 1 million tokens for input and output. Lower is better.
DeepInfra has the lowest estimated cost for 1,000 input and 500 output tokens at $0.0016.
Based on 1,000 input and 500 output tokens. Cost is shown in USD per 1 million tokens. Lower and further left is better; the single best cost/speed trade-off appears at full opacity.
3 providers have complete pricing and speed data. DeepInfra is closest to the ideal combination of lowest cost and fastest response.
Server-rendered comparison
Provider endpoints
Prompt and completion prices are effective USD per token as published by OpenRouter. Missing speed data never removes an endpoint.
Rank providers by
| Provider / route | Current pricing ↗ | Speed measurements | Endpoint details |
|---|---|---|---|
| Provider / route | Current pricing
| Speed measurements
| Endpoint details
18 parametersfrequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_p |
| Provider / route | Current pricing
| Speed measurements
| Endpoint details
17 parametersfrequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_p |
| Provider / route Venice venice/fp8 | Current pricing
| Speed measurements
| Endpoint details
13 parametersfrequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, reasoning_effort, response_format, stop, temperature, tool_choice, tools, top_k, top_p |
Endpoint data fetched .
Nemotron 3 Ultra deployment guide
How to choose a Nemotron 3 Ultra inference provider
Nemotron 3 Ultra is indexed here as a text model by NVIDIA, with 3 provider endpoints from DeepInfra, Together, Venice. The comparison preserves each exact OpenRouter routing tag so pricing and performance observations can be connected to the route an application would actually request.
Match the endpoint to the workload
The model publishes a 1M-token context window. It accepts text input and returns text output. 18 distinct supported parameters appear across the listed routes. Confirm limits on the specific endpoint rather than assuming every host exposes the same configuration.
Compare the real request economics
DeepInfra currently has the lowest estimated cost for the standard 1,000-input/500-output-token sample at $0.0016. Input-heavy and output-heavy applications can produce a different result, so review both per-million-token prices in the endpoint table.
Balance response start and generation speed
Together currently has the shortest estimated 500-token response at 7.48 seconds. DeepInfra is the single endpoint closest to the current ideal cost/speed combination. Recent observations can change, so validate finalists with your own prompts.