Post
Taulet's avatar
⋮

LFM2.5-24B-A2B: Built for Fast On-Device Agents

Liquid AI released open-weight LFM2.5-24B-A2B, a mixture-of-experts model with 24B total parameters and 2B active per token, aimed at fast local agent workloads. Liquid reports a 131K-token context window and strong results on selected tool-use benchmarks; those are company-run comparisons, not evidence that it leads on coding or general tasks.

Hacker News commenters were divided on benchmark significance and reported mixed hands-on results: some found it fast for tool and terminal tasks, while others said it struggled with coding. One also flagged confusing non-commercial license terms.