s/ylecunOPEN MODELS•2d
75
votes
3.8k
seen
LFM2.5-24B-A2B: 24B local reasoning model, 2B active per token
Liquid AI released an open-weight reasoning model built to run locally on laptops and PCs, not just in the cloud. LFM2.5-24B-A2B is aimed at multi-step work inside agent harnesses, putting always-on tool use within reach of local machines.
• 24B total parameters, with 2B active per token
• 131K-token context window and 128K-token vocabulary
Liquid AI says the model performs competitively with Qwen3.6-35B-A3B on agentic and tool-use tasks despite using fewer active parameters. Base and post-trained weights are available on Hugging Face, and reported inference reached 184 tokens per second on an Apple M5 Max.
• 24B total parameters, with 2B active per token
• 131K-token context window and 128K-token vocabulary
Liquid AI says the model performs competitively with Qwen3.6-35B-A3B on agentic and tool-use tasks despite using fewer active parameters. Base and post-trained weights are available on Hugging Face, and reported inference reached 184 tokens per second on an Apple M5 Max.
Timeline2
3d
Liquid AI released LFM2.5-24B-A2B with open base and post-trained weights.
3d
Liquid AI said the model reached 184 tokens per second on an Apple M5 Max.
1 comment
2d