Anthropic $70B projection 💰, Nano Banana 2 🍌, beyond standard LLMs 🧠

Anthropic projects API revenue of $3.8 billion this year, double OpenAI, and is targeting $20-26 billion in annual recurring revenue for 2026 ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR AI 2025-11-05 [Research] Connecting monitoring and AI to business outcomes (Sponsor) In The State of Observability 2025 report by Dynatrace, research findings show that 98% of ITOps leaders use AI for security compliance, but only 28% align observability data to business KPIs. As enterprises scale AI, observability is shifting from reactive monitoring to a strategic intelligence layer. Explore the full report >> 🚀 Headlines & Launches Anthropic projects $70B in revenue by 2028 (3 minute read) The company projects API revenue of $3.8 billion this year, double OpenAI, and is targeting $20-26 billion in annual recurring revenue for 2026. OpenAI is aiming for $100 billion in revenue by 2027, but expects to lose $115 billion before reaching profitability in 2029, compared to Anthropic, which is projected to be cash flow positive by 2028. Google is preparing Nano Banana 2 for the upcoming release (2 minute read) Google's upcoming GEMPIX2 model (Nano Banana 2) could be available as soon as next week. It is expected to be targeted at creators and professionals who rely on Gemini for image generation and AI-assisted creative workflows. The Nano Banana line has attracted considerable interest and contributed to a noticeable growth in Gemini's user base. OpenAI's Sora Video Generator App Finally Available on Android (2 minute read) Sora is now available on Android, making hyperrealistic video generation capabilities available to millions more users. The app quickly became a social media phenomenon when it was first released on iOS. It allows users to create realistic videos with audio and impressive physics. Android users will still need an invite code to actively use Sora's generation features. The launch is limited to select countries, including the US, Canada, Japan, Korea, Taiwan, Thailand, and Vietnam. 🧠 Deep Dives & Analysis Chaining ffmpeg with a Browser Agent (4 minute read) 100x included ffmpeg into its Chrome extension to make complex media processing a single, serverless, and stateless step in any workflow. It already had a browser agent for automation - ffmpeg became another 'tool call' for the agent. This enabled the agent to deal with complex editing tasks like mixing audio and video and adding text over video. This post gives readers a look at how the integration works. Beyond Standard LLMs (46 minute read) The largest and most capable open-weight large language models today are autoregressive decoder-style transformers. Some alternatives have popped up in recent years. While some are geared toward better efficiency, others are aimed at improving model performance. This article gives an overview of alternative models that exist, from text diffusion models to the most recent linear attention hybrid architectures. How We Built a Custom Vision LLM to Improve Document Processing at Grab (10 minute read) Grab developed a lightweight, specialized Vision LLM from the ground up to address the limitations of traditional Optical Character Recognition (OCR) systems. Traditional OCR systems struggle with the variety of document templates they have to process and often fall short in understanding South East Asian languages. Grab's model can effectively replace traditional OCR pipelines, opening new possibilities for document processing at scale. This post details how the company built its new model and also provides performance results. 🧑‍💻 Engineering & Research AI Agents Are Everywhere, but Are They Delivering? (Sponsor) AI agents are rapidly becoming business-critical, but enterprise success is far from guaranteed. In fact, a new Dataiku/Harris Poll survey of 800+ global data leaders found that 95% admit they cannot fully trace AI decisions for audit purposes. The "Global AI Confessions Report: Data Leaders Edition" uncovers the hidden risks, misalignments, and urgent corrections leaders need to turn agents from a liability into measurable business value. Discover the hidden risks, fragile foundations, and urgent corrections needed for AI to deliver value in this report. Code execution with MCP: Building more efficient agents (15 minute read) Code execution can enable agents to interact with MCP servers more efficiently. When too many servers are connected, tool definitions and results can consume excessive tokens, reducing agent efficiency. Many of the problems with MCP feel novel, but they have known solutions from software engineering. Code execution applies established patterns to agents, letting them use familiar programming constructs to interact with MCP servers more efficiently. Run Any LLM with a Single API: Introducing any-llm v1.0 (2 minute read) any-llm is an interface that can use any model, cloud or local, without requiring a rewrite of the stack every time. It makes it easy for developers to use any large language model without being locked into a single provider. any-llm decouples project logic from model provider with a stable and consistent API surface. Profiling with Cursor 2.0: The Missing Layer in AI Code Generation (5 minute read) Profiling is the best way to separate code that looks good from code that is actually good. The Pyroscope Performance Profiler is a Cursor extension that takes profiling data and paints it right on the code. It allows developers to look at multiple solutions generated by different agents and pick the best based on real numbers. Performance comparisons across deployments, trend analysis, and incident inspection can all be done inside the editor without going to a separate UI. Exploring a space-based, scalable AI infrastructure system design (5 minute read) Google will launch two satellites carrying its AI chips, called TPUs, in 2027, in hopes that in the mid-2030s space-based data centers will become more cost efficient since solar panels are up to 8x effective in orbit. Project Suncatcher envisions a constellation connected by optical inter-satellite links to power distributed ML tasks, like AI inference. 🎁 Miscellaneous Amazon Sends Perplexity a Cease and Desist Over Its AI Agents Shopping for You (6 minute read) Amazon has demanded that Perplexity stop its agentic web browser from buying products on behalf of humans. It claims that the Comet browser degrades the shopping and customer service experience that Amazon provides. Amazon has accused Perplexity of violating its terms of service and committing computer fraud by failing to disclose when an AI is shopping for a user. Perplexity responded by accusing Amazon of bullying and using litigation to stifle innovation. The Company Quietly Funneling Paywalled Articles to AI Developers (10 minute read) At the heart of AI web scraping is Common Crawl, a small nonprofit that supplies, and is funded by, major AI labs with internet data while flagrantly disregarding paywalls and removal requests. The organization has gathered millions of articles from The New York Times, The Atlantic, and other publishers (many of them now suing AI companies), but has avoided compliance by making its search tool return "no captures" for domains that actually contain thousands of articles. ⚡ Quick Links Getting agents to code is easy... but can you get them to follow your rules? (Sponsor) Most AI tools generate lines of code — but fail to understand your architecture, standards, and compliance rules. Discover how Tabnine bridges the gap Gemini's retention data is out, and it looks better than even ChatGPT! (1 minute read) Three-month retention is consistently over 90%. Coding Agents Are Outliers (10 minute read) The success of coding agents has created unrealistic expectations for AI agents in other domains. Shopify says AI traffic is up 7x since January, AI-driven orders are up 11x (3 minute read) Shopify is working with Perplexity and Microsoft to add other in-chat shopping experiences after the success of its integration with ChatGPT. Codemaps: Understand Code, Before You Vibe It (5 minute read) Cognition launched Windsurf Codemaps, which creates AI-annotated structured maps of codebases to address the onboarding costs of learning a new codebase or maintaining unfamiliar legacy code. Commitments on model deprecation and preservation (4 minute read) Anthropic will preserve weights of all publicly released Claude models for the company's lifetime and conduct exit interviews with models before retirement, documenting their preferences about future development. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/44fee20c/2 Track your referrals here. 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 or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! 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.