# A 2.6-Billion-Parameter Model Beats a 9-Billion Rival on Tool Use While Running on a Phone

Liquid AI's LFM2.5-2.6B scores 77.83 on ToolSandbox against 76.44 for Qwen3.5-9B, and a new benchmark released today argues on-device memory assistants are still badly under-tested.

- Published: 2026-08-05T05:32:18.568Z
- Canonical: https://polylog.news/ai/2026-08-05/a-2-6-billion-parameter-model-beats-a-9-billion-rival-on-too
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [Liquid AI](https://www.liquid.ai/blog/lfm2-5-2-6b), [Hugging Face](https://huggingface.co/blog/LiquidAI/lfm2-5-2-6b), [arXiv](https://arxiv.org/abs/2608.02613)

Liquid AI released LFM2.5-2.6B, a 2.6-billion-parameter model built for agent workloads that run entirely on the device. The Russian-language channel AI ML Big Data summarized the release as a local agent that plans, calls tools and execute…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-08-05/a-2-6-billion-parameter-model-beats-a-9-billion-rival-on-too (subscription information: https://polylog.news/pricing).