Menu

← Leaderboard

Nemotron-3-Nano-30B-A3B with Hermes Agent

Q8_0 (GGUF) on llama.cpp commit a37cc46. Tested

Task completion
83.3% (25/30)
Tool-call success
72.3%
Decode tok/s
56.9
Prefill tok/s
3166.4
Server memory
31.7 GB
Avg turns
4.7

Setup

Model
Nemotron-3-Nano-30B-A3B
Family / maker
Nemotron · NVIDIA
Parameters
30B (3B active)
Architecture
MoE
Weights
unsloth/Nemotron-3-Nano-30B-A3B-GGUF
Quant
Q8_0 (GGUF)
Engine
llama.cpp commit a37cc46
Hermes Agent
v0.22.4
Tested
2026-10-23
Context
64K (65,536 tokens); long-context task t09 at 128K or 192K
Memory tier
≤128 GB (measured)

Per-task results

  • t01: 3/3 · day run: 3/3
  • t02: 3/3 · day run: 3/3
  • t03: 3/3 · day run: 3/3
  • t04: 3/3 · day run: 3/3
  • t05: 3/3 · day run: 3/3
  • t06: 3/3 · day run: 3/3
  • t07: 1/3 · day run: 3/3
  • t08: 2/3 · day run: 2/3
  • t09: 1/3 (rerun 2026-10-23, 192K context) · day run: 0/3 (old harness, model never called)
  • t10: 3/3 · day run: 3/3

First number: night run 2026-10-23 (shown in the stats above).

Copy-paste config

1. Start the model server (llama.cpp)

Exact command used in our test. Download the GGUF from unsloth/Nemotron-3-Nano-30B-A3B-GGUF into ./models/.

llama-server --model ./models/Nemotron-3-Nano-30B-A3B-Q8_0.gguf --alias nemotron-3-nano-30b-a3b --host 127.0.0.1 --port 18080 --ctx-size 65536 --parallel 1 --n-gpu-layers 999 --flash-attn on --jinja

2. Point Hermes Agent at it

hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:18080/v1
hermes config set model.default nemotron-3-nano-30b-a3b

Other Nemotron setups