The Orchestrator › Models › NVIDIA
NVIDIA: Nemotron 3 Super
NVIDIA Nemotron 3 Super is a 120B-parameter open hybrid MoE model, activating just 12B parameters for maximum compute efficiency and accuracy in complex multi-agent applications. Built on a hybrid Mamba-Transformer...
Specifications
| Provider | NVIDIA |
| Model ID | nvidia/nemotron-3-super-120b-a12b |
| Context window | 262K tokens (262,144) |
| Modalities | text → text |
| Tokenizer | Other |
| Router sources | OpenRouter, TrustedRouter, Vercel AI Gateway |
| Released | 2026-03-11 |
| Tracked since | 2026-11-14 |
| Popularity | Not in the ranked set |
| Reasoning | On by default |
Pricing (per million tokens)
| Variant | Source | API ID | Input | Output | Context |
|---|---|---|---|---|---|
| standard | OpenRouter | nvidia/nemotron-3-super-120b-a12b | $0.075 | $0.4 | 262K |
| standard | TrustedRouter | nvidia/nemotron-3-super-120b-a12b | $0.079 | $0.422 | 262K |
| standard | Vercel AI Gateway | nvidia/nemotron-3-super-120b-a12b | $0.15 | $0.65 | 256K |
Recorded price history
| Date | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| 2026-11-14 | $0.15 | $0.65 | 256K |
| 2026-11-14 | $0.079 | $0.422 | 262K |
| 2026-11-14 | $0.075 | $0.4 | 262K |
Capabilities
Agentic tool useStructured outputsReasoning mode256K+ context windowControllable reasoningNo content storage (attested)Zero data retentionNo training on inputs (attested)
Trust flags come from TrustedRouter's per-provider attestations.
Supported parameters
frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p