Sponsored by
The World's Biggest Dev Event Hits Silicon Valley
WeAreDevelopers World Congress comes to San José, CA — September 23–25, 2026. 10,000+ developers, 500+ speakers, and the full software development lifecycle under one roof, in the heart of Silicon Valley.
Kelsey Hightower. Thomas Dohmke (fmr. CEO, GitHub). Christine Yen (CEO, Honeycomb). Mathias Biilmann (CEO, Netlify). Olivier Pomel (CEO, Datadog). The people actually building the tools you use every day — all on one stage.
AI, cloud, DevOps, security, architecture, and everything real builders ship with. Workshops, masterclasses, and the official congress party.
Hey everyone!
I’m doing a quick roll call to see who’s hanging out in our community these days. I'd love to learn more about what you do, the industry you're in, and what you're most curious about right now.
If you've got 60 seconds, tap the link below and let me know!
Today:
Anthropic’s New Opus 4.7 Trades Speed for "Human-Level" Reasoning and Honesty.
OpenAI’s New "Computer Use" Agent Can Run Your Mac for You.
Gemini 3.1 Brings Zero-Latency Voice Control
π 0.7 Pushes Robots Closer to Real-World Generalization
OpenAI Unveils GPT-Rosalind to Accelerate Drug Discovery
Well, Anthropic just dropped Claude 4.7 Opus, and it’s a beast.
According to Anthropic, Opus 4.7 improves on Opus 4.6 in complex long-running work, follows instructions more precisely, and is better at checking its own answers before responding. The model has substantially better vision with higher-resolution image understanding, and does better creative and professional work like interfaces, slides, and documents.
The "Human-Level" Coder: This model is designed for tasks that used to require constant human hand-holding. In early tests, it autonomously built a complex Rust engine from scratch—including the neural model and a browser demo and then verified its own work. It’s showing a level of "honesty" and self-correction we haven't seen before.
High-Def Vision: Its ability to process images has tripled. It can now handle high-resolution visuals (up to 2,576 pixels), making it much better at reading dense charts, complex diagrams, or tiny text in screenshots that used to get blurred out.
Literal Instruction Following: Anthropic warns that 4.7 is so precise that you might need to rewrite your old prompts. It no longer "fills in the blanks" or guesses what you mean; it follows your instructions exactly as written.
The Safety Trade-off: Interestingly, Anthropic admitted that Opus 4.7 is intentionally less powerful than their unreleased "Mythos" model. Why? Because Mythos is so good at cybersecurity that they’ve deemed it too risky for the general public right now. Opus 4.7 includes new "guardrails" specifically to block high-risk hacking requests.
OpenAI just gave Codex a massive promotion. It’s no longer just a tool for developers to write code; it’s becoming a central workspace for almost everything.
Computer Use: This is the big one. Codex can now "see" your Mac screen, move the cursor, click buttons, and type in apps just like a human. The coolest part? It can run these tasks in the background, so it doesn't hijack your mouse while you’re trying to work on something else.
Persistent Memory: It now remembers how you like to work. If you have specific coding conventions or a preferred "vibe" for your slide decks, Codex will remember those preferences across different sessions so you don’t have to repeat yourself.
Autonomous Workflows: You can now schedule Codex to handle tasks that take days or weeks. It can wake itself up, check on a process (like a long software test), and continue working without you needing to prompt it every five minutes.
Built-in Browser & Images: It now has an integrated browser where you can leave comments directly on web pages for the AI to fix, and it’s integrated with gpt-image-1.5 for generating mockups and assets on the fly.
Google just released Gemini 3.1 Flash-TTS, and it’s a game-changer for anyone who thinks AI voices sound too robotic. They’ve introduced a "vibe control" for speech.
Audio Tags: You can now use "audio tags" (like [whispers], [laughs], or [gasp]) directly in your text to control the performance. You can tell the AI to be "excited" or "cautious" mid-sentence using natural language commands.
Massive Scale: It supports over 70 languages and 30 conversational voices, making it one of the most versatile speech models out there.
Built-in Safety: To keep things responsible, all audio is watermarked with SynthID, an imperceptible mark that lets people verify if a voice is AI-generated.
🧠RESEARCH
Lyra 2.0 is an AI tool that creates massive, explorable 3D worlds from a single image. It solves common problems where AI forgets previously built areas or distorts graphics over time. By remembering older frames and self-correcting errors, it builds long, highly consistent 3D environments ready for real-time use.
Seedance 2.0 is a highly advanced AI model that generates video and audio using text, images, or sound as a starting point. It produces smooth 4-to-15-second clips at up to 720p resolution. It offers massive quality improvements over previous versions and includes a "Fast" mode for quicker, low-delay creation.
OccuBench is a new testing system that evaluates how well AI assistants perform in 100 real-world professions, like medical sorting or customs processing. It creates simulated workplaces with realistic errors and missing information. Testing shows that no single AI model is the best at handling every type of job.
📲SOCIAL MEDIA
🗞️MORE NEWS
Physical Intelligence Unveils Pi0.7 Physical Intelligence has released Pi0.7, a new brain for robots that makes them much more adaptable. Unlike older systems that only knew how to do one specific thing, this "foundation model" (a general-purpose AI base) allows robots to learn new, physical tasks—like folding laundry or clearing a table—much faster and with more human-like coordination.
OpenAI Launches GPT-Rosalind OpenAI has introduced GPT-Rosalind, a specialized version of its AI built specifically for biology and medicine. Named after the scientist who helped discover the shape of DNA, this tool helps researchers speed up "drug discovery" (the process of finding and testing new medicines) by analyzing massive amounts of scientific data to suggest new experiments.
Google and Pentagon Discuss Classified AI Deal Google is in talks with the U.S. military to allow its Gemini AI to be used for secret government work. To address ethical concerns, Google is trying to add rules to the contract that would prevent the AI from being used for "autonomous weapons" (machines that can make their own decisions to attack) or for spying on people without a human in charge.
Tesla's AI5 Chip Elon Musk announced that Tesla has finalized the blueprint for its newest self-driving computer chip and sent it to manufacturers. Creating this custom hardware helps Tesla power its future autonomous cars and robots without relying on outside suppliers. This move secures the massive computing power Tesla needs to build its next generation of technology.
Microsoft Takes Over Norway Data Center Microsoft is taking direct control over the massive computing centers in Norway that were previously managed by OpenAI. This strategic takeover consolidates the physical servers and hardware needed to run advanced AI models directly under Microsoft's roof. Ultimately, it gives Microsoft tighter control over the raw power driving the current technology boom.
The Next Evolution of the Agents SDK OpenAI updated its software toolkit to help developers build independent AI assistants that can safely inspect files, write code, and finish complex tasks. These assistants now operate inside secure digital workspaces, preventing them from accidentally breaking the user's main computer system.
Subagents in Gemini CLI Google added "subagents" to its developer tools, allowing the main AI to hand off specialized chores to smaller, focused AI assistants. Instead of the main AI trying to remember and do everything at once, these mini-assistants can research or test code in the background. This team-based approach drastically speeds up complex, time-consuming technical projects.
Novo Nordisk and OpenAI Partnership The healthcare company Novo Nordisk formed a major partnership with OpenAI to bring artificial intelligence into every step of creating new medicines. They plan to use AI to analyze complex biological data to find new treatments faster and to streamline their manufacturing operations. As part of the deal, OpenAI will also help train Novo Nordisk's global employees on how to properly use these new tools in their daily work.
Skills in Chrome Google launched a new browser feature that lets users save their most useful AI instructions as one-click shortcuts. Instead of typing the same complex request over and over, you can now reuse your customized commands to quickly summarize articles or compare products.
