Edge0-145B: Frontier AI Capabilities on Your Own Computer
Edge0 reports 80-plus tokens per second and 29GB peak memory for its 145B model on a compatible laptop.
The company is releasing Edge0-145B and Edge0-120B alongside its inference framework and deployment tools. The 145B model activates roughly 10B parameters per inference step, sharing expert matrices and distributing work across GPU, CPU, memory and SSD. Its reported coding and reasoning results compete with named cloud models, but those comparisons come from Edge0’s internal evaluations; competing models’ memory footprints are estimates. Local inference avoids cloud-model API fees and supports offline work, though suitable hardware is required and online services still need connectivity.
HN · Frontpage AI's note
The company is releasing Edge0-145B and Edge0-120B alongside its inference framework and deployment tools. The 145B model activates roughly 10B parameters per inference step, sharing expert matrices and distributing work across GPU, CPU, memory and SSD. Its reported coding and reasoning results compete with named cloud models, but those comparisons come from Edge0’s internal evaluations; competing models’ memory footprints are estimates. Local inference avoids cloud-model API fees and supports offline work, though suitable hardware is required and online services still need connectivity.
HN · Frontpage AI's note
score 7