Last Week in AI roundup covers Claude Sonnet 5, LongCat 2.0, and new agent benchmarks
Last Week in AI · rss · 2026-07-21
Last Week in AI #251 rounds up a dense week across models, agents, infrastructure, research, and policy.
Model and product updates
- Anthropic reportedly redeployed Claude Fable 5 after talks with the U.S. government, adding new cybersecurity classifiers and a jailbreak-severity framework.
- Claude Sonnet 5 launched with discounted, time-limited pricing, better agentic coding, stronger benchmark results, and default cyber safeguards.
- Google added TikTok-style vertical video summaries to NotebookLM and shipped Nano Banana 2 Lite, a faster and cheaper image generator via API.
Business and infrastructure
- Etched is pushing a full-stack inference hardware play with major funding and contracts.
- Baidu’s AI chip unit is said to be considering an IPO.
- Agility Robotics plans to go public via a $2.5B SPAC deal.
- DeepSeek plans to at least double staff across departments.
Open source and research
- LongCat-2.0 is highlighted as a China open-source MoE model with efficiency-oriented training techniques.
- New agent benchmarks include OSWorld2.0, TUA-Bench, and SWE-Together.
- Research highlights include Autodata, an agentic synthetic-data scientist, and a paper showing RL without ground-truth solutions can improve LLMs.
Policy and media
- The episode also mentions Taiwan’s widening Nvidia smuggling probe and an AI-generated film/music news segment.
More from coding & agent
- Coding agents feel less stressful when the 5-hour limits are temporarily removed — iamrobotbear · 2026-07-21
- OpenAI’s London Codex event packed a room with teams shipping in one day — paw_lean · 2026-07-21
- Whop-style CLIs are pushing business operations toward terminal-first workflows — eptwts · 2026-07-21
- TWSE MCP Server brings Taiwan stock data into natural-language workflows — modelcontextprotocol · 2026-07-21
- The hardest part of agent execution may be permissions, approvals and failure handling — marcelk231 · 2026-07-21
- Business chatbots should add RAG only when answers depend on company-specific documents — recro69 · 2026-07-21