Z.ai launched GLM-5.3 today, built on the same 743B base model as GLM-5.2 but with extreme post-training scaling — the model sets a new standard among open models for cybersecurity (beating Fable 5 and GPT-5.6 Sol on CyberGym) and sees a ~50% jump in coding capability. Open weights are staged for release in two weeks after safety evaluation.
DeepSeek officially launched V4-Pro yesterday (Aug 13), ending a four-month preview period. The 1.6T MoE model brings major agent upgrades, flexible reasoning effort levels, and native OpenAI Responses API support. Reuters reports DeepSeek is aggressively expanding hiring and compute to compete with domestic rivals.
The frontier AI leadership shuffle continues: Google DeepMind's new head Koray Kavukcuoglu inherits a race to catch OpenAI and Anthropic, with Gemini 3.7 Flash shipping as a down payment. Meanwhile, 1,367 frontier AI lab researchers signed an open letter warning of an arms race "beyond our ability to understand or control," asking the US government for international coordination to pace frontier development.
Same 743B base as GLM-5.2, all gains from post-training. Beats Fable 5 and GPT-5.6 Sol on CyberGym benchmark. ~50% coding capability jump. Open weights in two weeks after safety evaluation.
Formal release ending 4-month preview. 1.6T MoE, major agent upgrades, flexible reasoning effort, native Responses API support. Expanding hiring and compute aggressively.
Koray Kavukcuoglu replaces Demis Hassabis as SVP; Gemini 3.7 Flash stronger for coding, knowledge work, and web dev. Pressure mounts to ship Gemini 3.5 Pro.
Signers from OpenAI, Anthropic, Google DeepMind warn of capability acceleration "beyond our ability to understand or control." Ask US government for international pacing effort.
Both released open-weight models this week to compete with Chinese labs dominating the open-weight market. Joint push against "premature restrictions" on open models.
Selected posts from the edition, with links to the originals.
"Introducing GLM-5.3: Built to Code. Ready for Cyber Defense. Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model. A major leap in cybersecurity, setting a new standard among open models."
View post on X ↗"We're launching DeepSeek-V4-Pro today! 🚀 Major Agent upgrades with strong production gains! Flexible reasoning effort for V4-Pro & V4-Flash: low for simple tasks, high for daily Agent workflows, max for complex tasks. Native OpenAI Responses API support, optimized for Codex."
View post on X ↗"🧩 DeepSeek Harness v0.1 is now available in Developer Preview! We're opening it up to developers building agent harnesses worldwide and open-sourcing the codebase in MIT license. Powered by the Cordis meta-framework, DeepSeek Harness is an agent harness built around one core idea: Everything is a tool."
View post on X ↗"ChatGPT can now remember your activity across the apps and websites on your computer. With Computer History in the desktop app, future interactions feel more personalized and require less explanation."
View post on X ↗"Previewing Ultrafast mode: GPT-5.6 Sol at up to 14x the speed. Launching first in the OpenAI API to a select group of customers with expanded access to more businesses as capacity grows."
View post on X ↗"Gemini 3.7 Flash is here. It's stronger for coding, knowledge work, and web development."
View post on X ↗"🎵MiniMax-Music3: Next-Generation Open-Weights Production-Ready & Versatile Music Model"
View post on X ↗"A weird experiment I've been trying the last few weeks is having Claude take over day-to-day maintenance of our apps. Seeing early signs of life that this might be possible. The setup is straightforward: we have a Slack channel called proj-claude-maintains-apps. In it, Claude Tag runs a bunch of daily checks and maintenance tasks."
View post on X ↗"Introducing Bot Mode for Hermes Agent. Bot Mode is an alternative to sessions mode, where you have one chat with each agent profile, or 'bot'. These bots can be given jobs, descriptions, profile pics, and communicate with your other bots!"
View post on X ↗"Congratulations to the winners of the SpaceXAI Grokathon 2026. 1st Place: Nova taught Grok to reverse-engineer binaries into clean C. Started with 262KB GameBoy ROMs, ended up rebuilding a 1995 car's ECU runtime."
View post on X ↗