nomos-1-fp8

Nomos 1

We release Nomos 1, a specialization of Qwen/Qwen3-30B-A3B-Thinking-2507 for mathematical problem-solving and proof-writing in natural language. Nomos-1 was trained in collaboration with Hillclimb AI.

Nomos 1 is designed to be used with the Nomos Reasoning Harness, which we open-source concurrently.

On Putnam 2025, Nomos 1 scores 87/120 when wrapped in the Nomos reasoning harness. Under the same conditions, Qwen/Qwen3-30B-A3B-Thinking-2507 scores 24/120.

<img src="https://cdn-uploads.huggingface.co/production/uploads/6526f9d60d960cddf54e45cd/dkuxIB-Wg1hiJzSy7HkCr.png" height="60"/>

Usage

We recommend using Nomos-1 without a system prompt.

Hugging Face

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "NousResearch/nomos-1"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True,
)

messages = [
    {
        "role": "user",
        "content": (
            "Solve the following problem and show your reasoning:\n\n"
            "Let a, b, c be positive real numbers such that abc = 1. "
            "Prove that\n"
            "\\[\n"
            "  \\frac{1}{1+a} + \\frac{1}{1+b} + \\frac{1}{1+c} \\ge 1.\n"
            "\\]"
        ),
    },
]

inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=400,
        temperature=0.6,
        top_p=0.95,
        top_k=20,
        do_sample=True,
    )

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

SGLang

python -m sglang.launch_server \
  --model-path NousResearch/nomos-1 \
  --tp-size 8

vLLM

vllm serve \
  --model NousResearch/nomos-1 \
  --tensor-parallel-size 8

Acknowledgements

We would like to thank the following contributors:

Nous Research

Roger Jin, Jeffrey Quesnelle, Dakota Mahan, Chen Guang, Teknium

Hillclimb AI

Jun Park, Ibrakhim Ustelbay

Math experts

Samuel Kim, Miron Yurkevich, Adilet Zauytkhan, Rinat Amankos, Alex Andreyev, Damir Nurlanov, Abuzer Abuov

Other

massiveaxe

Citation

@misc{nomos_model2025,
  title        = {Nomos Model},
  author       = {Jin, Roger and Quesnelle, Jeffrey and Mahan, Dakota and Guang, Chen and Teknium, Ryan and Park, Jun and Ustelbay, Ibrakhim and Kim, Samuel and Yurkevich, Miron and Zauytkhan, Adilet and Amankos, Rinat and Andreyev, Alex and Nurlanov, Damir and Abuov, Abuzer and massiveaxe, Askar},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/NousResearch/nomos-1}},
  note         = {Model release},
}