Six Recent Small Open Models for On-Device AI: From PII Redaction to Video Understanding
vanstriendaniel · x · 2026-08-07
The author highlights several recently released open-weight small models, noting that many AI tasks can now run entirely locally or on-device.
These models cover various practical use cases:
- PII Redaction: OpenAI privacy-filter (1.5B) for browser-based data sanitization.
- Speech-to-Text: NVIDIA Nemotron-3.5 ASR (0.6B) with streaming support.
- Document Parsing: OvisOCR2 (0.9B).
- Agent Loops: LiquidAI LFM2.5 (2.6B).
- Content Moderation: Mistral Shieldstral (3B) with custom policy support.
- Video Understanding: Marlin-2B.
The author suggests that the next major challenge will be effectively routing and orchestrate these models locally.
Related event: Open-Source Small Models Enable Fully Local AI Execution(2 posts)→
More from Infra
- MotherDuck Lets AI Agents Self-Provision Data Warehouses Without Signup — JeremyCMorgan · 2026-08-07
- GitHub Outage Exposes Agent Bottleneck: Centralized Control Planes Limit Automation — amu4biz · 2026-08-07
- DeepSeek API Pricing Dispute: Can Rented GPUs Actually Cover Costs? — t4a8945 · 2026-08-07
- Open-Weight Small Models Enable Fully Local AI Tasks from ASR to Agents — vanstriendaniel · 2026-08-07
- SpaceX Generates Up to $50M/MW in Compute, Crushing Neoclouds — shaunmmaguire · 2026-08-07
- DeepInfra Processes Over 500B Tokens Daily Amid Surging Inference Demand — patricious · 2026-08-07