Nemotron 3.5 Bolt

NVIDIA's open 74B MoE model (9B active) packed into 24GB for local agents on RTX Spark; released Oct 15, 2026.

NVIDIA · Open-weight/community models · model

Published in Feed

Overview

Nemotron 3.5 Bolt 74B is an open mixture-of-experts model from NVIDIA with 74B total and 9B active parameters, released October 15, 2026 for agentic AI on RTX Spark laptops. NVIDIA fits it into 24GB using 2-bit quantization with quantization-aware training, and says the release includes training data, weights, and recipes. It runs on llama.cpp, vLLM, Ollama, Unsloth, and LM Studio, and is reachable from the Hermes and OpenClaw agents. A smaller Nemotron 3.5 Bolt E4B (4GB) for NVIDIA ACE game characters is scheduled for October 23, 2026.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • 74B total / 9B active parameter mixture-of-experts design
  • 24GB memory footprint via 2-bit quantization-aware training
  • Weights, training data, and recipes released as open source (license name not stated in the source)
  • Launch support in llama.cpp, vLLM, Ollama, Unsloth, and LM Studio
  • Directly accessible from the Hermes and OpenClaw agents
  • Nemotron 3.5 Bolt E4B (4GB) variant for NVIDIA ACE game characters due Oct 23, 2026

Use cases

  • Persistent local agents for coding, research, and monitoring on a laptop
  • Privacy-sensitive workloads that must stay on-device
  • Hybrid local/cloud agent setups routed through NVIDIA PAIR
  • Conversational NPCs in games (E4B variant)

Pricing

Free open weights; exact license name not confirmed. Running it locally requires hardware with about 24GB of memory.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
15 Oct 2026
Record updated
16 Oct 2026
Record last checked
16 Oct 2026

Explore the Feed

Other entries filed under open-weight/community models; inclusion does not imply equivalent features.

  • Muse Spark Open

    Meta's open-weight 816B-parameter MoE model (42B active, 1M context) for coding and long-horizon agent work, released Oct 15, 2026.

  • Jais 3

    Six open Arabic-focused language models from 900M to 375B parameters with 512K context, plus Jais Speech for Arabic recognition.

  • kettei

    Sqwish Labs' decision-model family returning answer probabilities in one pass, with open 3.7B and 900M models and a fine-tuning API (OneHot).

Browse the Feed