A Guth Labs publication ↗
Back to News

Edge0 releases large local-agent models, with platform and licensing limits worth reading

Edge0 releases large local-agent models, with platform and licensing limits worth reading

Edge0 has published 145B and 120B mixture-of-experts models derived from Qwen3.8-Flash-Next. The project's release log dates the announcement November 13; its indexed blog carries November 12. Exact release time and the reason for that date difference were not established. Official release log

Publication update, November 15, 2026: the current README lists Linux and Windows with NVIDIA GPUs/CUDA 13, plus macOS Apple Silicon with Metal for the 145B and 120B Q3_K models. The earlier unpublished draft described a platform table without Mac support; that description has been superseded. This is documented support, not a hands-on compatibility test. Current platform table

The 145B model card reports a 40.81 GiB Q3_K checkpoint. Its headline speed is 121.2 tokens per second for agent decoding on an RTX 5090 D with 24 GB VRAM. A constrained 16 GiB VRAM plus 20 GiB host-memory experiment is explicitly a test budget, not minimum requirements. These are developer measurements, not our reproduction. Model card

The weights are not simply Apache-licensed because the framework is. The model card names Edge0 Community Model License 1.0 plus applicable upstream Qwen terms. It says commercial model services and independent commercial coding or office assistants require a separate written Edge0 license. Developers should examine the full terms for their intended use. Model-card licensing section

Analysis

The interesting development is a concrete local-serving option, not a universal “145B on any laptop” claim. For agent builders, the evaluation should include tool correctness, latency under realistic context, host-memory use and product licensing. A headline parameter count or decoding figure cannot settle those questions.

Evidence limits

Official GitHub and Hugging Face pages directly read during the 2026-11-13T07:53:33Z run. No software installed, weights downloaded or benchmark reproduced. Secondary AI Brief coverage was indexed but failed direct retrieval; no independent validation claimed. Generic Hugging Face library suggestions are not treated as tested compatibility.

Sources

  • Primary cited: https://github.com/Edge0-AI/Edge0 ; release note dated 2026-11-13, current mutable documentation.
  • Primary cited: https://huggingface.co/Edge0/Edge0-145B-q3k ; current model card, precise publication/update timestamp unknown.
  • Additional background, not relied on for performance: https://edge0.ai/blog/introducing-edge0-145b-frontier-ai-capabilities-on-your-own-computer ; indexed date 2026-11-12; direct retrieval failed.
  • Independent reporting lead, not cited evidence: https://dailyaibrief.com/news/edge0-releases-145b-agent-model-for-one-consumer-gpu-0ju8456l ; indexed 2026-11-13; direct retrieval failed.

Sources and citations

Revision history

Published .