NVIDIA Nemotron 3.5 Bolt 74B for local agents
A 74B mixture-of-experts model, 9B active, packed into 24GB for local agents on RTX Spark.
Official source (opens in a new tab)
NVIDIA's 13 October 2026 RTX Spark post says Nemotron 3.5 Bolt 74B-A9B is available Thursday, 15 October 2026. It is a 74 billion parameter mixture-of-experts model with 9 billion active parameters, packed into 24GB with 2-bit quantization and quantization-aware training, and built for local agents on RTX Spark. NVIDIA says it is faster than most popular dense models and smarter than most popular mixture-of-experts models with a similar memory footprint. It can be run through llama.cpp, vLLM, Ollama, Unsloth, and LM Studio, and through agents including Hermes and OpenClaw. NVIDIA says the model is fully open source, including training data, weights, and recipes.
The same post says NVIDIA PAIR, a free open-source personal AI router, moved out of beta on 13 October, with llama.cpp support. A smaller Nemotron 3.5 Bolt E4B 4GB model is a separate release dated Friday, 23 October.
Who it's for: people running local agents on RTX Spark hardware, not a cloud coding assistant swap.
Fits: local agent work next to Best Agent Builder and Best for Vibe Coding. It does not change the order of those lists.