Nemotron 3.5 Bolt 74B-A9B

Name: Nemotron 3.5 Bolt 74B-A9B
Author: NVIDIA

NVIDIA · Nemotron 3.5 · released 2026-10-12 · data checked 2026-11-01

Nemotron 3.5 Bolt 74B-A9B is a 74B-parameter Mixture-of-Experts model (9B active per token) from NVIDIA (United States), released 2026-10-12 under the OpenMDW-1.1 license. It scores 73.3 on the Small LM Index, #1 of 6 large models. It needs about 46.9 GB of memory at 4-bit with an 8K-token context.

A 2-bit build fits on a single 24 GB GPU (Terminal-Bench 2.1: 51.0, GPQA: 86.4).

Small LM Index

73.3

#1 of 6 scored large models

Likely 71.9–74.8 · from 2 of 6 core benchmarks

Parameters

74B

9B active · MoE

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

262K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best large model on that benchmark. Sources: 4 reported by NVIDIA. How we label sources.

GPQA Diamond Lab-reported Knowledge88.8

Humanity's Last Exam Lab-reported Reasoning28.7

Terminal-Bench Lab-reported Agentic · Terminal-Bench 2.1, BF1653.9

AA-LCR Lab-reported Long context67.4

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

MMLU-Pro≈ 86

AIME≈ 95

LiveCodeBench≈ 84

SWE-bench Verified≈ 69

Specifications

Lab: NVIDIA (US · United States)
Architecture: Mixture-of-Experts
Total parameters: 74B
Active parameters: 9B
Size class: Large
Context window: 262K tokens
Input modalities: text, image
Reasoning: Hybrid (thinking on/off)
License: OpenMDW-1.1
Openness: Open weights
Memory to run: ~46.9 GB at 4-bit, 8K context (more sizes)
Hugging Face: 7.7K downloads (30 days) · 52 likes

Weights on Hugging Face ↗

Compare with

Nemotron 3.5 Bolt 74B-A9B vs DeepSeek-R1-Distill-Llama-70B (70.6B, index 50.1)
Nemotron 3.5 Bolt 74B-A9B vs Llama 3.3 70B (70.6B, index 40.8)
Nemotron 3.5 Bolt 74B-A9B vs Kolibri-1 (78.1B, index 68.3)
Nemotron 3.5 Bolt 74B-A9B vs K2-V2 Instruct (70B, index 57.9)
Nemotron 3.5 Bolt 74B-A9B vs Qwen3-Next 80B-A3B Thinking (80B, index 63.3)

Ranked in

Sources

https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16

Something wrong here? Report it.

Run it locally

Uses unsloth/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-GGUF (UD-Q4_K_XL). Swap the tag for another size: BF16, IQ2_XXS, Q8_0, UD-IQ1_M, UD-IQ1_S, UD-IQ2_M, UD-IQ2_XXS, UD-IQ3_XXS….

Ollama

ollama run hf.co/unsloth/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-GGUF:UD-Q4_K_XL

llama.cpp

llama-server -hf unsloth/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-GGUF:UD-Q4_K_XL

Then open http://localhost:8080 for a chat window.

vLLM (GPU server)

vllm serve nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	256K context
16-bit F16	150 GB	150 GB	152 GB
8-bit Q8_0	81 GB	81 GB	83 GB
4-bit Q4_K_M	47 GB	47 GB	49 GB

Check it against your GPU, Mac or phone →