ModelsPublished All news from that day
Liquid AI lists LFM2.5-24B-A2B, an on-device model, and reports its own LiveClaw score of 48.09.
The Hugging Face model card lists 2.45B active parameters, the lfm1.0 licence and a 131,072-token context; the benchmark comparison is Liquid AI's own figure.
Check our sources · 4 facts from 1 sourceLiquid AI's model card on Hugging Face describes LFM2.5-24B-A2B as a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device.
The card lists 24.0B total parameters, 2.45B active parameters, 64 experts per mixture-of-experts block with top-4 routing, a training budget of 36 trillion tokens and a context length of 131,072 tokens. It lists the licence as lfm1.0 and offers both a pre-trained base model and a version post-trained for agentic workloads.
The card reports, as Liquid AI's own figures, a score of 48.09 on LiveClaw against 38.19 for Qwen3.6-35B-A3B.
Underlined sentences link to the facts we checked. Dotted words open a short explainer.
Fact 2 of 4 · from Liquid AI (Hugging Face)
64 experts per MoE block with top-4 routing, a 131,072-token context window, and a training budget of 36 trillion tokens
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against this source.
1 LiquidAI/LFM2.5-24B-A2B model card
Open the source archived copy-
Liquid AI's model card on Hugging Face describes LFM2.5-24B-A2B as a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device.
cite24B-parameter reasoning model with only 2B active parameters
-
The card lists 24.0B total parameters, 2.45B active parameters, 64 experts per mixture-of-experts block with top-4 routing, a training budget of 36 trillion tokens and a context length of 131,072 tokens.
cite64 experts per MoE block with top-4 routing, a 131,072-token context window, and a training budget of 36 trillion tokens
-
It lists the licence as lfm1.0 and offers both a pre-trained base model and a version post-trained for agentic workloads.
citelfm1.0
-
The card reports, as Liquid AI's own figures, a score of 48.09 on LiveClaw against 38.19 for Qwen3.6-35B-A3B.
cite48.09 ... 38.19 for Qwen3.6-35B-A3B
Every sentence above is checked against this source. Each fact has its own address you can cite.
Source 1 · Liquid AI (Hugging Face)
LiquidAI/LFM2.5-24B-A2B model card
Liquid AI's model card on Hugging Face describes LFM2.5-24B-A2B as a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device.
24B-parameter reasoning model with only 2B active parameters
The card lists 24.0B total parameters, 2.45B active parameters, 64 experts per mixture-of-experts block with top-4 routing, a training budget of 36 trillion tokens and a context length of 131,072 tokens.
64 experts per MoE block with top-4 routing, a 131,072-token context window, and a training budget of 36 trillion tokens
It lists the licence as lfm1.0 and offers both a pre-trained base model and a version post-trained for agentic workloads.
lfm1.0
Show all 4 factsShow fewer facts
The card reports, as Liquid AI's own figures, a score of 48.09 on LiveClaw against 38.19 for Qwen3.6-35B-A3B.
48.09 ... 38.19 for Qwen3.6-35B-A3B
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep