The official blog and model card confirm the weights, architecture, runtime integrations, vendor benchmarks and custom-license restriction.
Liquid AI releases the LFM2.5-24B-A2B reasoning model
2026-10-14Frontier watchTRIAL
Liquid AI released base and post-trained weights for LFM2.5-24B-A2B, a 24B-parameter mixture-of-experts model that activates 2B parameters per token. It targets reasoning and tool use across local devices and servers, with day-one support in llama.cpp, MLX, vLLM, SGLang and ONNX. The custom license permits many commercial uses but excludes entities with at least $10 million in annual revenue unless separately licensed.
What this means for you
The model offers a testable route to stronger local agent inference without activating all 24B parameters, but vendor benchmarks and the revenue-gated license limit immediate production use. Compare it on your own tool schemas, latency and memory budget before committing.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Architecture | 24B total parameters, 2B active per token |
|---|---|
| Artifacts | public base and post-trained weights; no hosted inference provider on the model card at review time |
| Runtime Support | llama.cpp, MLX, vLLM, SGLang and ONNX |
| Vendor Results | Liquid AI reports wins or ties on five of eight agent and tool-use benchmarks; coding and knowledge results are less consistent |
| Vendor Speed | 184 tok/s on M5 Max, 114 tok/s on Ryzen AI Max+ 395 and over 21K output tok/s on B300 at high concurrency |
| License | LFM Open License v1.0; separate license required for entities with annual revenue of at least USD 10 million |
| Time Basis | Hugging Face blog feed timestamp 2026-10-13T17:05:48Z |
Technical impact 4
Engineering value 4
Adoption signal 2
Maturity 3
Verification cost 3
Risk 4