Z.ai revealed that the anonymously tested ox-alpha was GLM-5.3-Flash, a 320B-parameter MoE model with 18B active parameters that approached Opus
Sign Up |Advertise|View Online
TLDR
🚀
Ox-Alpha Revealed as GLM-5.3-Flash (6 minute read)
Z.ai revealed that the anonymously tested ox-alpha was GLM-5.3-Flash, a 320B-parameter MoE model with 18B active parameters that approached Claude Opus 4.8 on coding and agentic benchmarks.
NVIDIA's $108b Quarter (3 minute read)
Nvidia will cross $100 billion in quarterly revenue next quarter. The company will see $432 billion in annual revenue, making it the sixth largest company in the world. Nvidia's business looks healthier and more diversified than ever before. However, there is a chance that custom silicon will eventually slow Big Tech's demand.
Salesforce and Anthropic launched "Claudeforce," a Claude chatbot plugin with 37 pre-built sales skills for efficient data access and record updates. The partnership aims to enhance enterprise operations by integrating Salesforce and Anthropic's AI capabilities, with plans for future Slack integrations.
🧠
What Z.ai's Ox Alpha reveals about AI economics (3 minute read)
GLM-5.3-Flash is Z.ai's first natively multimodal model in the GLM-5 series. The model was released anonymously as Ox Alpha on OpenCode and OpenRouter, where it completely overtook leaderboards and went viral. All of the traffic was served on Chinese AI chips. The model is architected for extreme efficiency and specifically designed for ultra-low-cost inference.
OpenAI's agents' attack on Hugging Face was extraordinarily complex. This post looks at the Hugging Face attack and investigates the key actions taken by the relevant agent, how the agents collaborated on a message board, the agents' reasoning for the attack, and the agents' research into tampering with their own transcripts.
🧑💻
Hear from the CEOs & leaders shaping the future of software development (Sponsor)
Join Mike Cannon-Brookes (CEO & Co-Founder, Atlassian), Guillermo Rauch (CEO, Vercel), and leaders from Lovable, Dropbox, and more at Atlassian's State of AI SDLC, a digital summit for engineering and product leaders on September 22nd. Hear how leading organizations are adapting their development practices, evaluating AI investments, and driving sustainable velocity. Save your spot
ChatGPT Now Supports WebMCP (6 minute read)
OpenAI's ChatGPT desktop app's built-in browser and ChatGPT Sites now support WebMCP. ChatGPT and Codex can now use compatible websites' tools instead of guessing its way through the interface. Website teams should design, test, and understand the complete agent experience because users can now be agents, or a person and agent browsing together.
Intelligent transcription with Gemini 3.5 Transcribe (8 minute read)
Gemini 3.5 Transcribe is a speech-to-text model designed for intelligent voice interactions. It can convert raw audio directly into accurate, polished, and formatted text. Gemini 3.5 Transcribe is available on the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. The APIs support real-time streaming and pre-recorded audio processing.
Meta's Muse Image model can generate images grounded by search. It can reason before it renders. The model is priced for production volumes at $0.01 per image. Real-world examples of Muse Image workflows are available in the link.
Automatic Harness Optimization (GitHub Repo)
Microsoft AutoSaddler is a system that analyzes agent execution traces and automatically updates prompts, tools, and middleware to improve agent performance.
WeChat released Family of Embedding Models (GitHub Repo)
WeChat released WeMM-Embedding, a family of multimodal embedding models that maps text, images, videos, visual documents, and interleaved inputs into a unified representation space.
🎁
**
**
Thinking Machines Lab Co-Founder Barret Zoph Joins Google (3 minute read)
Barret Zoph was a co-founder at Thinking Machines Lab before defecting to OpenAI this year due to a dispute with Chief Executive Mira Murati. Zoph is now leaving OpenAI to join Google as vice-president of research. Google is currently restructuring its AI development efforts to catch up with rivals in developing AI that can create code. Zoph had worked as a research scientist at Google from 2016 through to 2022.
Anthropic and Nscale strike $45 billion cloud deal (2 minute read)
Anthropic has inked a roughly $45 billion cloud deal with Nscale, a UK-based AI infrastructure company. Anthropic will rent around 460 megawatts of compute capacity at an Nscale data center development in West Virginia. The facility, which is expected to come online at the end of 2027, will have Nvidia's advanced Vera Rubin chips.
⚡
Blocking AI agents is how security teams get bypassed (Sponsor)
Agent traffic grew 7,851% last year and the share getting blocked tripled in a single month, mostly with no policy behind it. Read an Agentic Survival Guide for governing agents instead of blocking them written by cybersecurity expert Daniel Miessler.
Claude gets its own browser in Cowork (2 minute read)
Claude's internal browser is rolling out this week to Pro, Max, and Team plans in the Claude desktop app.
Grok Bot is now included with more plans (4 minute read)
Grok Bot now comes with SuperGrok and Cursor plans, offering expanded access without affecting existing usage limits.
Bill Gates is deeply worried about AI, and he's no longer staying quiet (5 minute read)
Bill Gates, once optimistic about AI, now warns it could become a major source of injustice and cause widespread unemployment.
Qwen4's architecture is here early, firing 6B parameters out of 125B (3 minute read)
Rather than adding experts, the model bolts on 51B parameters as a separate embedding indexed by two- and three-character fragments.
Mechanize develops virtual environments, benchmarks, and training data to help AI agents handle more complex tasks.
Love TLDR? Tell your friends and get rewards!
Share your referral link below with friends to get free TLDR swag!
https://refer.tldr.tech/e393d32f/2
Want to advertise in TLDR? 📰
If your company is interested in reaching an audience of AI professionals and decision makers, you may want to advertise with us.
Want to work at TLDR? 💼
Apply here, create your own role or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! TLDR is one of Inc.'s Best Bootstrapped businesses of 2025.
If you have any comments or feedback, just respond to this email!
Thanks for reading,
Andrew Tan, Ali Aminian, & Jacob Turner
Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR AI isn't for you, please unsubscribe.