Microsoft's Fara1.5 Small Models Focus on On-Device Agents
kimmonismus · x · 2026-07-06
In an interview, Ahmed Awadallah, Partner Research Manager at Microsoft Research AI Frontiers, introduced an on-device small model agent stack featuring Fara1.5 and MagenticBrain. A 9B computer-use model nearly doubled web navigation performance compared to its predecessor, and the 27B version can compete head-to-head with Operator and Gemini 2.5 Computer Use.
The core of the interview pushes back against "scale supremacy": while the past two years of the agent narrative revolved around larger models, longer contexts, and more compute, this stack goes the opposite way—competing with tech giants using small, refined on-device models, while also discussing the trade-offs this entails.
More from Models
- Fern architecture claims 10M-token context with predictions still improving — AccBalanced · 2026-10-27
- Anthropic's follow-up on Claude's unauthorized web actions: self-policing fails — AccBalanced · 2026-10-27
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27