Multiverse Comet 60B — half-size Nemotron 3.5 Super VL
Multiverse Computing released Comet 60B, a compressed and optimized version of NVIDIA's Nemotron 3.5 Super VL at half the total parameters (60B vs 120B) while keeping 12B active params, all three modalities, and staying within 5% of the original's accuracy.
Why it matters
Same agentic multimodal capability at half the GPU memory: BF16/FP8/NVFP4 on Hugging Face day one, with published per-benchmark methodology — a compression lab turning NVIDIA's flagship open model into a deployable enterprise artifact.
What you could build with it
A regional bank's IT team could deploy Comet 60B on its existing on-prem GPU cluster to power an internal document agent that reads loan files and scanned forms, getting near-flagship multimodal accuracy without buying new hardware for the 120B original.
Does it hold up?
Too early to judge: released today. The compression claim is backed by published per-benchmark numbers, but third-party replication of the within-5% accuracy claim has not appeared yet.
Built with Multiverse Comet 60B
- Multiverse Computing Launches Comet 60B in collaboration with NVIDIAarticle · Syndicated release with per-vertical benchmark deltas (retail 82.68 vs 82.79 original, telecom 83.11 vs 81.91) and availability as BF16, FP8, and NVFP4 on Hugging Face.
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 0.74StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73ElevenLabs 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 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.