Interfaze open-weights MoA model for OCR, speech and GUI grounding on one 80GB GPU

charles_irl · x · 2026-10-06

Startup Interfaze released interfaze-1-lite, an open-weight (Apache-2.0) mixture-of-architectures model aimed at "deterministic" developer workloads. A vision-language reasoning core dispatches specialist perception architectures and composes structured answers, running entirely on a single 80GB GPU with no external services.

Capabilities include document understanding (text, layout, tables from images, PDFs up to 50 pages, Word files — with bounding boxes and confidence per word), speech transcription with timestamps and diarization (95-minute audio in 90 seconds), open-vocabulary object detection and GUI grounding for computer-use agents, schema-constrained structured output, 160+ language translation, time-series forecasting, and guardrails. Weights are on Hugging Face.

Original post →

More from coding & agent

coding & agent channel →