All models

LFM2.5-24B-A2B

A small downloadable model built to run on modest hardware.

Last updated October 16, 2026.

Costs here use the prices AI companies charge software developers, so they show relative cost, not your monthly bill. See what ChatGPT, Claude and Gemini charge you.

Made by Liquid AI. Free to download.

What it costs

1,000-word essay: under 1 cent
100 chat messages: $0.07 (about 7 cents)
50-page PDF summary: under 1 cent

Based on list prices and the assumptions on How we work. Your real cost will differ.

At a glance

Smartness: Not rated yet (our banding of the Artificial Analysis score, where available)
Reads at once: about 190 pages (rule of thumb)
Understands: text
Free to download and run yourself: Yes
Free to use: Not free for developers. Free app plans

Best for

running on modest hardware

Watch out for

Reads fewer pages at once, and has no smartness score yet.

Frequently asked

How much does LFM2.5-24B-A2B cost to use?
A 1,000-word essay costs about under 1 cent; summarizing a 50-page PDF about under 1 cent. These are developer (API) prices, not an app subscription.

What is LFM2.5-24B-A2B best for?
A small downloadable model built to run on modest hardware.

How much can LFM2.5-24B-A2B read at once?
About 190 pages (a rule of thumb: 128,000 tokens at roughly 500 words a page).

Can I download LFM2.5-24B-A2B for free?
Yes, the model files are published, though running a large one needs powerful hardware.

Under the hood

Maker	Liquid AI
Price per 1M input tokens	$0.050
Price per 1M output tokens	$0.70
Context size	128K tokens (128,000)
Artificial Analysis Intelligence Index	Not reported (source, via OpenRouter)
Open weights	Yes (Hugging Face: LiquidAI/LFM2.5-24B-A2B)
Speed	Not measured yet
Listed	2026-10-13

Maker's description (technical)

LFM2.5-24B-A2B is a mixture-of-experts reasoning model from Liquid AI with 24B total and 2.45B active parameters. It always reasons before answering and is built for agents that call tools. Liquid...

View on OpenRouter. Our verdict above is editorial, not benchmarked by us.