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

Fields

id

sources/lfm2-5-24b-a2b-blog

URL

https://www.liquid.ai/blog/lfm2-5-24b-a2b

Kind

blog

Accessed

2026-10-19

notes

Read https://www.liquid.ai/blog/lfm2-5-24b-a2b for the benchmark figures on models/lfm2-5-24b-a2b. As recorded by the reader: Values cross-checked between the model card benchmark figure and the official blog's Table 1 (identical). The blog row is labelled 'SWE-Bench Verified (OpenCode)', i.e. SWE-bench Verified run in the OpenCode harness; reported here as SWE-bench Verified. Other reported benchmarks (AutomationBench, BFCLv4, ToolSandbox, LiveClaw, PinchBench, tau3-Bench Banking, WebArena Verified, AA-Omniscience, HLE, IFBench, Multi-IF, SciCode) are outside the allowed list.

Notes

Read https://www.liquid.ai/blog/lfm2-5-24b-a2b for the benchmark figures on models/lfm2-5-24b-a2b. As recorded by the reader: Values cross-checked between the model card benchmark figure and the official blog's Table 1 (identical). The blog row is labelled 'SWE-Bench Verified (OpenCode)', i.e. SWE-bench Verified run in the OpenCode harness; reported here as SWE-bench Verified. Other reported benchmarks (AutomationBench, BFCLv4, ToolSandbox, LiveClaw, PinchBench, tau3-Bench Banking, WebArena Verified, AA-Omniscience, HLE, IFBench, Multi-IF, SciCode) are outside the allowed list.

Cross-references

This record references no other records.