This episode of the AI Daily Brief covers the latest model releases and tools from major AI companies, hosted by Nathan Labenz. The show is sponsored by KPMG, Blitzy, Zencoder, and Drata.

The discussion begins with clarification around OpenAI's rumored SPUD model restrictions, then moves to Perplexity's explosive revenue growth following their Computer launch. The episode covers GitHub's infrastructure strain from the surge in AI-generated code commits, reaching unprecedented levels.

The main focus shifts to new model releases including Meta's MuseSpark from their Superintelligence Lab, China's GLM 5.1 open source model, and Anthropic's Claude Managed Agents platform. The episode concludes with Google's new Notebooks feature for Gemini, which integrates project management capabilities across their product suite.

OpenAI SPUD Model Confusion and Perplexity's Growth

Initial reports about OpenAI restricting their SPUD model release due to cybersecurity risks were incorrect - Dan Shipper clarified that OpenAI has a separate cyber product in testing, not SPUD itself

Perplexity's revenue effectively doubled in a single quarter after launching Computer in February, reaching 450 million ARR with 100 million monthly active users and tens of thousands of enterprise clients

The finance sector shows particular enthusiasm for Perplexity Computer, with Geiger Capital noting "AI demand is still accelerating" and "nobody is ready for the compute we need"

GitHub Infrastructure Strain from AI Coding Surge

GitHub is experiencing unprecedented growth with 275 million commits per week, putting them on track for 14 billion commits annually compared to 1 billion total commits last year

Commits to public repos from AI-generated code have increased 25x in the past six months, reaching 2.5 million last week according to GitHub COO Kyle Daigle

The surge is revealing infrastructure limits, with more frequent outages and API quota issues prompting GitHub to push "incredibly hard on more CPUs, scaling services, and strengthening core features"

Meta's MuseSpark: First Model from Superintelligence Lab

MuseSpark represents Meta's return to frontier models after over a year, scoring 52.4 on SweetBench Pro and 42.8 on Humanity's Last Exam, positioning it competitively but not leading the pack

The model excels in visual reasoning with an 86.4 score on Charvik's reasoning benchmark, beating Gemini 3.1 Pro by six points for a state-of-the-art result

Mark Zuckerberg positioned MuseSpark for "personal superintelligence" use cases including "visual understanding, health, social content, shopping, games" rather than enterprise coding applications

The model will operate in three modes: instant (no reasoning), thinking mode (enables reasoning), and contemplating mode (deep research) - though contemplating mode won't be available at launch

GLM 5.1: Open Source Model Beats Western Rivals

GLM 5.1 achieved 58.4 on SuiteBench Pro, becoming the first open source model to surpass GPT-4 (57.7) and Claude Opus (57.3) on coding benchmarks

The 754 billion parameter model demonstrated autonomous capabilities by spending "eight hours autonomously building a Linux desktop using a self-review loop" without human intervention

ZAI leader Lou noted the progression in agent capabilities: "Agents could do about 20 steps by the end of last year. GLM 5.1 can do 1700 right now"

The model was trained entirely on Huawei chips, demonstrating Chinese hardware capabilities while maintaining only a months-long gap behind US frontier models

Anthropic's Claude Managed Agents Platform

Claude Managed Agents provides "everything you need to build and deploy agents at scale" with production infrastructure allowing developers to "go from prototype to launch in days"

The platform includes an agent harness, sandboxed environments, autonomous cloud execution for hours, and permission management - eliminating the need for dedicated infrastructure engineering teams

Common usage patterns include event-triggered agents ("a system flags a bug and a managed agent writes the patch"), scheduled tasks, and fire-and-forget operations via Slack or Teams

Current limitation noted by users: "someone still has to tune the prompt every Friday and act on the brief by 9 a.m. Monday" - the agent writes but operators must still act on outputs

Google's Notebooks Feature for Gemini

Google introduced Notebooks in Gemini as "personal knowledge bases shared across Google products" that integrate NotebookLM-style resource management directly into the Gemini app

The feature allows users to organize resources, documents, and context for particular tasks while building custom instruction sets for different projects

Josh Woodward from Google positioned this as more than basic projects: "Most AI chatbots give you basic projects. Gemini just built you a second brain"

Resources Mentioned

A Practical Guide to Law School in the Age of AI Success Strategies for the Modern Law Student (The Modern Law Student Series)

regarding whether political considerations can drive federal procurement. Charlie Bullock, a senior research fellow at the Institute for Law and AI, told the information he was unsurprised by the resu

The Idea of You A Novel

he managed agent to do a task via Slack or Teams, and long horizon tasks like Andre Karpathy's auto-research idea. Now it's early, but some of the first experiments seem to validate some of those patt