# Nemotron 3 Super 120B on the Cotera agent benchmark

> The open-weights 5/5 — swept the matrix for $0.61, an order of magnitude cheaper than the frontier models.

Source: https://cotera.co/benchmarks/models/nemotron-3-super-120b

---

NVIDIA's Nemotron 3 Super, a 120B open MoE (12B active), run through OpenRouter. The mid-tier of NVIDIA's Nemotron 3 line, above Nano and below the 550B Ultra.

**Score:** 5/5 benchmarks passed · **Cost across the matrix:** $0.610

## Strengths
- Best value in the suite: a clean 5/5 at $0.61 — cheaper than every other model that swept the matrix.
- Real tool use and schema-conforming output across all five benchmarks — coding, CX complaints, Crunchbase, Apollo pricing, and Reddit tool-ranking.
- Concludes efficiently once pointed at the task: a "pick one thread, then stop and answer" system prompt kept it from over-searching on the open-ended Reddit benchmark.

## Watch-outs
- Needs the OpenRouter tool-call middleware: like the rest of the Nemotron family it sometimes emits tool calls as plain text and sends numeric args as strings, both of which have to be repaired for the agent loop to work.
- Short-horizon agent tasks only (2-11 tool calls here). No data on long multi-hour chains or adversarial inputs.