# Liquid AI Releases LFM2.5-1.2B-Thinking Model With On-Device Reasoning Under 1GB

> While the headlines chase ever larger frontier models, Liquid AI is pushing hard in the other direction. Its new LFM2.5-1.2B-Thinking is a 1.2 billion parameter reasoning model small enough to fit in about 900 megabytes of memory, which means it can run entirely on a modern phone at roughly 82 tokens per second on the device's neural chip. No cloud, no round trip, no data leaving the handset. The interesting claim is that going small did not mean giving up on reasoning. Through a multi-stage training recipe built around thinking tokens, and by tackling the failure mode where tiny models get stuck in repetitive doom loops, Liquid says it lifted the model's MATH-500 score from 63 to 88. It reports matching or beating Qwen3-1.7B on reasoning while using about 40 percent fewer parameters. For anyone who cares about private, offline, low-latency AI, this is the direction that matters. A capable reasoner that lives on your phone changes what assistants can do when there is no connection and no willingness to ship your data to someone else's server.

_Section: [Daily AI Updates](https://www.wortins.com/daily-ai) · Source: Liquid AI · Published Saturday, August 1, 2026_

## Wortins' read

While the headlines chase ever larger frontier models, Liquid AI is pushing hard in the other direction. Its new LFM2.5-1.2B-Thinking is a 1.2 billion parameter reasoning model small enough to fit in about 900 megabytes of memory, which means it can run entirely on a modern phone at roughly 82 tokens per second on the device's neural chip. No cloud, no round trip, no data leaving the handset. The interesting claim is that going small did not mean giving up on reasoning. Through a multi-stage training recipe built around thinking tokens, and by tackling the failure mode where tiny models get stuck in repetitive doom loops, Liquid says it lifted the model's MATH-500 score from 63 to 88. It reports matching or beating Qwen3-1.7B on reasoning while using about 40 percent fewer parameters. For anyone who cares about private, offline, low-latency AI, this is the direction that matters. A capable reasoner that lives on your phone changes what assistants can do when there is no connection and no willingness to ship your data to someone else's server.

## Source

[Read the full story at Liquid AI](https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb)

## Related coverage

- [Letter: the Senate disaster management subcommittee, led by Sen. Josh Hawley, is probing OpenAI's handling of the Hugging Face breach, calling it "reckless" (Axios)](https://www.wortins.com/story/letter-the-senate-disaster-management-subcommittee-led-by-se-da54156c) — [Techmeme](https://www.techmeme.com/260910/p11#a260910p11)
- [Researchers used AI to build a WeChat worm that spreads through phone calls](https://www.wortins.com/story/researchers-used-ai-to-build-a-wechat-worm-that-spreads-thro-8bf3f469) — [The Next Web](https://thenextweb.com/news/wechat-worm-ai-calif-tencent-zero-click)
- [Amazon Prime Video’s new AI tech matches lips to dubbed audio](https://www.wortins.com/story/amazon-prime-video-s-new-ai-tech-matches-lips-to-dubbed-audi-a93c2b22) — [The Verge](https://www.theverge.com/tech/991809/amazon-prime-video-ai-lip-sync-dubbing)
- [Two years ago, Meta killed CrowdTangle. Can a new AI tool fill the void?](https://www.wortins.com/story/two-years-ago-meta-killed-crowdtangle-can-a-new-ai-tool-fill-5e67f291) — [Nieman Lab](https://www.niemanlab.org/2026/09/two-years-ago-meta-killed-crowdtangle-can-a-new-ai-tool-fill-the-void/)
- [IBM and NASA release an open-source lunar foundation model](https://www.wortins.com/story/ibm-and-nasa-release-an-open-source-lunar-foundation-model-0522b9e8) — [The Next Web](https://thenextweb.com/news/nasa-ibm-lunar-foundation-model-open-source)
- [Top AI spenders cut per-employee costs by nearly 10 percent in August](https://www.wortins.com/story/top-ai-spenders-cut-per-employee-costs-by-nearly-10-percent--b31ffb0b) — [The Decoder](https://the-decoder.com/top-ai-spenders-cut-per-employee-costs-by-nearly-10-percent-in-august/)

---

_Curated and written by [Wortins](https://www.wortins.com) — The daily AI briefing. Every story links to its original source; the "Wortins read" on each is our own original analysis. [About Wortins & our editorial approach](https://www.wortins.com/about)._