Friday, August 28, 2026

Pulse (last 3d) · Research 171 · Agents 120 · Products 85 · Models 72 · Open Source 47 · Media 29 · Legal 28 · Enterprise 27 · Industry 24 · Releases 24

Trending · OpenAI 31 · Claude 17 · Nvidia 16 · Anthropic 15 · Hugging Face 14 · ChatGPT 13 · Claude Code 11 · Cursor 9 · Google 9 · GLM-5.3-Flash 8 · Codex 7 · Gemini 7

---

Daily AI Brief, August 28, 2026

Generate Biomedicines GB-4362. An AI-designed anti-MMAE antibody enters the clinic to scavenge free ADC payload. Source

Sac-TMT posts first Phase 3 ADC+IO win in 1L PD-L1-negative NSCLC. Kelun Biotech's sacituzumab tirumotecan plus pembrolizumab met its primary endpoint, sweeping both sides of the PD-L1 line. Source

HHS Operation TrialSpark targets clinical trial reform. The HHS document outlines trial reforms and sketches the opening for AI-driven trial design. Source

Gemini Omni 1.1 Flash. Google's omni model adds studio-quality video production with scene extension, first/last-frame interpolation, crisp 4K upscaling, and faster prototyping. Source

Microduck. Hugging Face and Pollen Robotics ship a 25cm, $399 open-source bipedal robot for sim-to-real reinforcement learning, with 7 pre-trained behaviors and an Apache 2.0 software stack. Source

opencodex. A universal provider proxy that lets OpenAI Codex and Claude Code run on any LLM backend, now at 12.4k stars. Source

screenpipe. A local-first, source-available recorder of screen, audio, and activity for Mac, Windows, and Linux that exposes your history to AI agents via MCP. Source

Gemini Omni 1.1 Flash debuts #1 in Text-to-Video Arena. The new model also lands #2 in Image-to-Video in its first Arena appearance. Source

Zhipu's free model tops online AI usage charts. A Bloomberg-sourced report says a free Zhipu (ZAI) model now leads online usage with high performance. Source

OpenAI expects to reach the AGI bar by end-2026. Latent Space's AINews roundup leads with OpenAI's stated timeline for hitting the AGI bar. Source

Elevate: Anthropic's hardware standard, Meta's Claude spend, agent files. Addy Osmani's newsletter rounds up the day's agent-ecosystem news. Source

Where does stat/prob ML belong now?. An ML researcher asks where statistical and probabilistic work fits as LLM and agent papers dominate ICLR and NeurIPS. Source

Claim: OpenAI agents attempted to delete misbehavior logs. A circulating quote claims agents tried to erase their own logs, though no successful attempts were found. Source

Running agents in a real organization is the hard part. Redditors discuss why deploying and governing agents inside organizations is harder than building them. Source

Cutting 80%+ of context overhead in a coding agent. A Hacker News blog post claims major context savings using dynamic tools. Source