H Company releases Holotron-12B computer-use agent
h company released holotron-12b, a 12b parameter computer-use agent model post-trained from nvidia's nemotron-nano-2 vl. the hybrid ssm architecture achieves 2x higher throughput than holo2-8b on webvoyager benchmarks, scaling to 8.9k tokens/s on a single h100 gpu with 100 concurrent workers.
H Company released Holotron-12B, a 12 billion parameter multimodal agent optimized for computer use. Built on NVIDIA’s Nemotron-Nano-2 VL architecture, the model uses a hybrid State-Space Model (SSM) and attention mechanism to minimize VRAM usage and avoid quadratic computation costs. The final checkpoint processed approximately 14 billion tokens on proprietary localization and navigation data.
In WebVoyager benchmarks running on a single H100 GPU with vLLM v0.14.1, Holotron-12B achieved over 2x higher throughput than the previous Holo2-8B model. Token throughput scaled linearly to 8,900 tokens/s at 100 concurrent workers, while accuracy on WebVoyager increased from 35.1% to 80.5%.