LLM Releases
← Catalog

Edge0-145B

Available
AutoArk (Edge0)Open weights

Open-weight on-device MoE model from AutoArk's Edge0 team, released Nov 12 2026 with its inference framework. 145B total parameters with about 10B active per step; groups of 64 experts share two large matrices to cut memory. Edge0 reports 29GB peak memory and 80+ tokens/s decode on a compatible laptop (e.g. MacBook Pro), 128K-256K usable context, and internal scores of 64.6 DeepSWE 1.1, 86.52 Terminal-Bench 2.1 and 90.4 GPQA-Diamond. Hugging Face checkpoints (NVFP4 and q3k) list Qwen3.8-Flash-Next as the base model. Released under the Edge0 Community Model License 1.0.

Specifications

License
Open weights · Edge0 Community Model License 1.0
Weights
Downloadable
Architecture
Mixture-of-Experts
Parameters
145B · 10B active
Context window
— tokens
Max output
—
Knowledge cutoff
—
Price (in / out, $/M)
—
Modalities
TextCode

Benchmarks

No benchmark scores recorded yet. Spotted some? Submit a correction.

Vendor-reported figures are claims until independently verified. See methodology.