NVIDIA RTX Spark Bolt 74B: 74B-parameter reasoning model that fits a 24GB laptop
NVIDIA's RTX Spark line launches Friday with Bolt 74B, a 74-billion-parameter reasoning model that occupies 24GB at inference using sparse activation and quantization-aware training (based on ICLR 2026 research), with weights, training data and recipes to be published openly.
Why it matters
Evidence that frontier-quality reasoning no longer requires cloud GPUs: a dense-equivalent 74B model in 24GB, shipping through llama.cpp, vLLM, Ollama, Unsloth and LM Studio with full open weights and training recipes.
What you could build with it
A field-services company could run the model fully offline on technicians' laptops, letting it draft repair plans from equipment manuals and job photos with no cloud round trip. The result: AI-assisted repairs in dead zones and client sites with strict data policies.
Does it hold up?
Too early to judge: launches Friday. The sparse-activation plus QAT recipe is grounded in published ICLR 2026 research, but no independent inference benchmarks or third-party builds exist yet.
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71JetBrains Mellum 2.1: 12B MoE coding model trained for agentsJetBrains released Mellum 2.1, a 12B MoE model (2.5B active parameters, 128K context) under Apache 2.0, trained mainly…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.