gekro
GitHub LinkedIn
News

AI News

Liquid AI Releases On-Device Agentic Model; Microsoft Open-Sources Unit-Test Agent

Liquid AI released LFM2.5-2.6B, a 2.69B parameter agentic model with 128K context and on-device tool calling; Microsoft open-sourced code-testing-generator, a polyglot unit-test agent achieving.

1 min read 4 sources

Liquid AI released LFM2.5-2.6B, a 2.69B parameter model designed for on-device agentic inference with tool calling and multi-step task completion capabilities (MarkTechPost). The model features 22 double-gated short convolution blocks paired with 8 GQA blocks across 30 layers, handles 131,072 tokens of context, and achieves 220 tokens per second decoding performance. The architecture demonstrates efficiency gains viable for edge deployment and local inference scenarios.

Microsoft open-sourced code-testing-generator, a polyglot unit-test agent in the MIT-licensed dotnet/skills repository that reads repositories before writing, detecting language, test framework, and build conventions, then plans and writes unit tests (MarkTechPost). The agent achieved 92.1% task completion versus 78.9% for stock Copilot. Cloudflare released Kitesurf, an agent-first web browser built for V8 isolates on Cloudflare Workers with no Chromium dependency, optimized for machine-readable content extraction and agent automation rather than human UX features (MarkTechPost). Prime Intellect open-sourced Prime Agent, a coding harness using a Recursive Language Model abstraction that turns sub-agent calls into functions inside persistent IPython kernels, allowing agents to edit their own prompts and skills (MarkTechPost).

Compiled automatically from the linked sources and published without manual editing - a neutral summary of third-party reporting, for information only. Every claim links to its origin. Not original reporting.