Liquid AI ships 2.6B phone-ready reasoning model trained for agentic workflows
JosephJacks_ · x · 2026-09-03
Liquid AI released LFM2.5-2.6B, a compact 2.6B-parameter reasoning model small enough to run on a phone and specifically trained for agentic workflows: planning, tool calling, and multi-step tasks.
Key points:
- Post-training included agentic RL inside real agent harnesses like OpenClaw and Hermes Agent
- 66K context; suited for agent workflows, data extraction, RAG, and long-context processing
- Not recommended for agentic coding or knowledge-heavy tasks
- Live on OpenRouter for free (liquid/lfm-2.5-2.6b:free), measuring 0.71s P50 latency and 132 tok/s throughput
- Caveat: prompts and outputs may be retained to train Liquid models
More from coding & agent
- Developer asks: is LangChain still worth it vs rolling your own agent harness? — curious_vii · 2026-09-03
- Inference Engineering Is Just a Recipe: vLLM/SGLang, Replicas, Cache-Aware Routing — GabGarrett · 2026-09-03
- Developer accidentally built an entire agent factory with Fable 5.1 — 0xkarasy · 2026-09-03
- Microsoft adds Fabric data agents to Foundry agents via Fabric IQ (preview) — adnan_hashmi · 2026-09-03
- Databricks pitches agent-native data infrastructure, Lakebase Postgres at VLDB 2026 — matei_zaharia · 2026-09-03
- doodlestein ships a comprehensive web app review skill after months of debugging — doodlestein · 2026-09-03