How Edge0 launched Edge0-145B and Edge0-120B on X
Open-weight 100B-class models that run offline on your laptop
What Edge0 launched

Introducing Edge0-145B and Edge0-120B, @Edge0_AI’s first device-native 100B-class foundation models with open weights. Cloud-level intelligence on your laptop. Free, offline, and fast enough for real vibe coding. - Smart: beats Claude Opus 4.6 on Terminal-Bench 2.1 (86.5% vs. https://t.co/KaqYWrQztm
How LaunchVideo.com rated the launch
LaunchVideo.com rating
About the launch
Edge0 has open-sourced Edge0-145B and Edge0-120B, sparse mixture-of-experts models for developers who want coding and agent work to run on their own machine. The 145B model is derived from Qwen3.8-Flash-Next, trained for 4-bit local execution, and keeps about 10 billion of its 145 billion parameters active on each token. The company reports a 29GB peak memory footprint, decode above 80 tokens per second, and prefill above 1,100 tokens per second, with weights on Hugging Face and inference code on GitHub. This is its first 100B-class release, after the Edge0-35B phone model it open-sourced in September. Repo docs for the current engine target Linux and Windows with an NVIDIA GPU, CUDA 13, at least 16GB of GPU memory, and 20GB of system memory, and they also publish throughput on an Apple M2 Ultra. In Edge0's internal evaluations, Edge0-145B scores 86.52 on Terminal-Bench 2.1, above the 78.2 it lists for Claude Opus 4.6 (Max), and 64.6 on DeepSWE 1.1, with 90.4 on GPQA-Diamond and 80.6 on IFBench. The company says context runs from 128K to 256K tokens and that a technical report will include the full evaluation setup. For operators, that means local coding sessions with no cloud token charge, and with source code that can stay on the machine. Samuel Zeng, founder and CEO, holds a PhD in artificial intelligence and was named to MIT Technology Review's Innovators Under 35. Edge0's stated focus is models built for the memory, power, and latency limits of hardware people already use.
Tags
Questions
Frequently asked questions
The Edge0 launch post reached 734K views on X, with 1.6K likes, 226 reposts, 169 comments, 1.6K bookmarks.
Edge0 launched on X on November 12, 2026 with the message "Open-weight 100B-class models that run offline on your laptop".
Edge0 has open-sourced Edge0-145B and Edge0-120B, sparse mixture-of-experts models for developers who want coding and agent work to run on their own machine.
LaunchVideo.com rated the Edge0 launch 2.60 out of 5: hook 2/5, editing / creativity 1/5, copy 4/5, sentiment of launch 3/5, distribution strategy 3/5.






