TLDR AI 2025-07-09

How does OpenAI manage billing for millions of users across thousands of enterprise contracts?

In 2021, OpenAI moved from homegrown, manual billing to scalable infrastructure powered by Metronome. This blog covers key learnings, including:

→ How OpenAI's approach to pricing and billing enables them to navigate high-cost, high-risk innovation

→ OpenAI's strategies for token-based pricing, prepaid credit models, and agile pricing updates—and how those support rapid product launches

→ Building precision into pricing decisions and treating billing as a product experience

Read the full story on the Metronome blog

Replit and Microsoft have partnered to deliver natural-language-based enterprise app development, integrating with Azure services to let business users create and deploy production-ready software without coding experience.

Mistral is negotiating to raise $1 billion from investors like Abu Dhabi's MGX fund and French lenders such as Bpifrance SACA. The French AI startup, known for its Le Chat chatbot and open-weight LLMs, has already raised $1.19 billion, reaching a $6.51 billion valuation. Mistral is collaborating with MGX and Nvidia to build Europe's largest AI data center, aligned with the UAE's €50 billion AI investment in France.

Gemini Nano is almost here for all Chrome users. This post is a rewrite of Google's documents for the integration. Developers will be most interested in the Prompt API, which is the most flexible and open-ended API. The post covers how to set up the feature, important things to be aware of, and pitfalls to look out for.

A Harvard Law graduate who argued before the Supreme Court fed his case briefs into Claude 4 Opus and had it answer the same questions the Justices posed to him. The AI delivered what he called an "outstanding oral argument" with coherent answers and clever responses he hadn't considered, leading him to conclude that AI lawyers could soon outperform even top human advocates at oral argument.

Join a team making transformative AI breakthroughs with unmatched resources and scale. From our AGI SF Lab to our Amazon Nova foundation models, we're combining pioneering research with real-world impact, backed by Amazon's industry-leading infrastructure and commitment to long-term innovation.

Explore AGI Careers at Amazon >>Join our Amazon AGI SF Lab >>

A fully open-source, LlamaCloud-backed alternative to NotebookLM, this project uses LlamaCloud for document processing, OpenAI for content generation, and ElevenLabs for voice synthesis.

Hugging Face's SmolLM3 is a fully open 3B-parameter language model that supports six languages, strong reasoning capabilities, and long-context processing. It targets high performance in the small model segment.

An open-source memory layer for AI agents that enables long-term, personalized interactions by efficiently storing and retrieving user context across sessions, reducing token costs and improving response accuracy.

OLMo 2 is a family of fully-open language models. OLMo 2 32B is the most capable and largest model - it is the first full-open model to outperform GPT3.5-Turbo and GPT-4o mini on a popular suite of benchmarks. OLMo 2 7B and 13B are competitive with open-weight models from Meta and Mistral on English academic benchmarks. OLMo 2 1B, the smallest member of the model family, surpasses peer models like Gemma 3 1B and Llama 3.2 1B.

In the few states where it's legal, in-person psychedelic therapy can cost thousands of dollars per session, gated by lengthy waiting lists. As a cheaper, faster alternative, users are turning to AI tripsitters. Companies are following suit too: Mindbloom, for example, is already integrating AI into its ketamine programs. Experts doubt that AI can provide the emotional attunement they consider essential for safe psychedelic experiences.

OpenAI has reportedly poached high-level engineers from Tesla, xAI, X, Meta, and other companies. The engineers include Tesla's Vice President of Software Engineering David Lau, X and xAI's head of infrastructure engineering Uday Ruddarraju, fellow xAI infrastructure engineer Mike Dalton, and Angela Fan, an AI researcher from Meta. This article looks at the new hires and their backgrounds and specialties.

You'll probably like our flagship newsletter. It's all about tech, science, and programming.

Same quick format. Still free.

Subscribe now.

LangChain is an AI infrastructure startup that provides tools for building and monitoring large language model-powered applications.

After alleging that DeepSeek improperly copied its models using distillation techniques, OpenAI implemented "information tenting" policies, biometric access controls, and offline isolation of proprietary technology.

The companies are launching the National Academy of AI Instruction with teachers' unions as schools struggle with AI's role in education.