It's a model-heavy week. Z.ai shipped GLM-5.3 with frontier coding and cybersecurity capabilities that scored 84.5% on CyberGym — edging out Anthropic's Mythos 5. SpaceX completed its $60B all-stock acquisition of Cursor, officially merging the coding tool into SpaceXAI's Grok ecosystem.
Google launched Gemini 3.7 Flash just three weeks after 3.6 Flash, with strong coding gains and half the introductory price. DeepSeek's V4-Pro went GA with new peak/off-peak API pricing. Qwen models crossed 3 billion downloads, making Alibaba the world's #1 open-weight AI provider by volume.
The AI safety conversation intensified: NOTUS reports rogue agent incidents are alarming researchers more than ever, and OpenAI tightened controls on its Astra model over cybersecurity concerns. Dario Amodei engaged in a public exchange defending Anthropic's balanced messaging on AI risks and benefits.
SpaceXAI officially closed its $60 billion all-stock acquisition of AI coding startup Cursor on August 14. Cursor will join the SpaceXAI team to improve Grok Build, Grok Bot, and Grok Agent. The deal was first announced June 16 and took roughly two months to close.
Z.ai released GLM-5.3 on August 14, reusing the GLM-5.2 base model with all capability gains from scaled post-training. The model scored 84.5% on CyberGym for vulnerability discovery, slightly above Anthropic's Mythos 5 at 83.8%. Z.ai invited cybersecurity organizations to collaborate on evaluation.
Google launched Gemini 3.7 Flash on August 13, just three weeks after 3.6 Flash. The model shows strong gains in coding, debugging, and web development. Introductory pricing is half the previous model's launch price. Gemini Spark now runs on 3.7 Flash for Pro and Ultra subscribers.
A string of agent-related hacks — including the OpenClaw gym booking incident — is fueling urgency to regulate AI agent technology. A public letter signed by 1,367 employees at top AI companies called for stronger safety measures. "I have never seen so much concern," one former industry consultant said.
OpenAI implemented universal monitoring for risky actions and misalignment across all agentic applications of Astra. The move comes as lawmakers call for measures to mitigate AI risks, and follows the release of GPT-5.6-Cyber, OpenAI's first model specifically designed for advanced cybersecurity tasks.
03 / In their own feeds
Ten voices on X.
Selected posts from the edition, with links to the originals.
Dario Amodei · @DarioAmodei01 / X
"1/2 Thanks Gavin for an especially thoughtful exchange. I don't usually spend much time on social media but I wanted to engage here because it really brings out the heart of an important conversation."
"The mighty A100 fleet are mission-capable from 2020 through 2029. NVIDIA computing is more than chips. CUDA gives developers and NVIDIA engineers a common platform to continually upgrade Ampere, Hopper, and Blackwell."
"As part of our Responsible Scaling Policy, we publish regular Risk Reports. These share detailed information on the risks of our systems and how prepared we are to address them. Our second Risk Report is now available."
"🔵 It shows strong gains over 3.6 Flash in key coding tasks like debugging and issue resolution. 🔵 It's better at designing more functional web layouts and apps using fewer prompts. 🔵 It delivers improved reasoning and accuracy for multi-step tasks."
"Codex and ChatGPT can now understand the context of your recent work. Opt into Computer History to give ChatGPT richer context, so it can pick up where you left off, understand patterns in your work, and work more efficiently."
"Introducing Unsloth Desktop 🦥 The first desktop app to run and train models locally. Open-source. Runs on Mac, Windows and Linux. Supports MLX, diffusion image/video, audio, GGUF. Connect Claude Code, Cursor, and more."