Edge0 has open-sourced 145B and 120B AI models designed to run locally on compatible laptops, with Edge0-145B reporting 29GB peak memory use and speeds above 80 tokens per second. This matters because demanding coding and agent tasks can run offline, keeping data on the device and avoiding cloud token or API fees.
The company says Edge0-145B matches or exceeds selected cloud models on coding, terminal tasks, reasoning, and instruction-following benchmarks, though several comparisons use internal evaluations and estimated memory figures. It can plan tasks, write and run code, debug failures, inspect visual results, and operate tools for projects such as 3D scenes, games, drones, and PCB designs.
The release combines model weights, an inference framework, and deployment tools for macOS, Windows, and Linux, including Apple M-series and NVIDIA hardware. Its architecture shares parameters across sparse experts and coordinates GPU, CPU, system memory, and SSD access to make the large model practical on personal computers.
Local AI still depends on suitable hardware, and cloud services remain useful for current information, cross-device memory, and devices that cannot run large models. The next step is for developers to download, evaluate, fine-tune, or integrate the models through Hugging Face, GitHub, or the Edge0 app.
Takeaways
The release's main technical distinction is not just model size: sparse expert sharing and coordinated GPU, CPU, system-memory, and SSD execution are presented as the mechanisms that make a 145B model practical on a laptop.
The strongest evidence is concentrated in company-reported benchmarks and demonstrations, so independent testing of memory, speed, and task reliability will determine whether the claimed cloud-level capability generalizes.
Article timeline
Build this article’s timeline
Extract dated events, key actors, and open questions using the same timeline analysis available in Signal.
Edge0 has open-sourced 145B and 120B AI models designed to run locally on compatible laptops, with Edge0-145B reporting 29GB peak memory use and speeds above 80 tokens per second.
🤖 AI prefill
📅When
July 2025 and September 2026
🤖 AI prefill
📍WhereWhere did this occur?
💡Why
Centers on Edge0-145B, on-device AI, and local inference
🤖 AI prefill
Login to answer and earn XP for each gap you fill.
Manipulation Score: LowHow much this article uses emotional framing vs. straight reporting
Loaded Language2225%
Absolutist1815%
Threat/Panic825%
Us vs Them515%
Engagement Bait3020%
Manipulation score look right?
Factual Claims (5)
lowtechnology
Edge0 has open-sourced Edge0-145B and Edge0-120B for local use on compatible laptops.
Today, we're open-sourcing Edge0-145B and Edge0-120B, together with our unified inference framework and deployment tools.
mediumtechnology
Edge0-145B reportedly uses 29GB peak memory and decodes at more than 80 tokens per second.
A 145B model runs locally on a compatible laptop, with reported peak memory of 29GB and decoding above 80 tokens/s.
mediumtechnology
The company says Edge0-145B is competitive with selected frontier cloud models on coding, reasoning, terminal, and instruction-following benchmarks.
Edge0-145B is competitive with specific frontier cloud models across coding, reasoning, knowledge, and instruction following
mediumtechnology
Edge0-145B can plan tasks, write and run code, debug errors, inspect visual results, and operate tools for complex projects.
The model plans, writes and runs code, debugs errors, and checks results visually for projects such as 3D scenes, games, hardware programs, and PCB designs.
lowtechnology
The release includes model weights, an inference framework, and deployment tools supporting macOS, Windows, Linux, Apple M-series chips, and NVIDIA platforms.
For Edge0-145B, day-one support covers macOS, Windows, and Linux, including Apple M-series chips and NVIDIA platforms
Comments
Login to view and post comments.