# Nemotron 3 Nano 30B on the Cotera agent benchmark

> The small one. 0/5 — cheap, but never assembled a passing answer.

Source: https://cotera.co/benchmarks/models/nemotron-3-nano-30b

---

NVIDIA's Nemotron 3 Nano, a 30B open MoE (3B active), run through OpenRouter. The smallest and fastest of the Nemotron 3 line.

**Score:** 0/5 benchmarks passed · **Cost across the matrix:** $0.411

## Strengths
- Cheapest Nemotron to run at $0.41 for the matrix.
- Got genuinely close on Crunchbase — it pulled the right funding total and correctly identified the Series D round.

## Watch-outs
- Went 0/5. The failures were a mix: right data in the wrong format, output degenerating into repetition, and looping past the tool-round cap without ever writing a final answer.
- On CX it ran 16 tool calls and was cut off before it could answer — the small model explores without converging.
- Like the rest of the family it needs the string-to-number argument coercion and the text-tool-call repair just to execute tools at all.