🧠 Jev turns AI calls into decisions and Opus 5.5 gets cheaper

⚡ aiBytes_ #11

Hello, builders 👋

Welcome back to your weekly dose of AI Bytes. Last week belonged to Jev, and I'm excited about it.

Jev is the first model from TypeSafe AI, the new company of Diogo Almeida, who helped build ChatGPT. An LLM predicts the next word. Jev makes a decision: it is a classifier that answers in well under a second, for a fraction of an LLM call's cost.

How it works: hand Jev a customer's support email and your list of teams, then ask "Is this urgent?", "Which team owns it?" and "How upset is the customer, 1 to 5?" in one call. You get a probability for every option instead of a paragraph to parse.

Diogo's talk on why they built Jev is a must-watch 🎥

This week: the posts that went viral on X, five launches from Clueso MCP to Arcjet, five trending repos, and four HN deep stories worth your weekend.

🧬 Claude discovers a novel enzyme system - The system carries CRISPR-like repeats, and Anthropic says humans are still in the loop in its biology lab. Story of the week (Anthropic)

🚀 Anthropic releases Claude Opus 5.5 - The first model in the Claude 5.5 family performs at the level of Claude Fable 5.1 on most tasks and costs 40% less to run than Opus 5. (Anthropic)

☀️ OpenAI launches GPT-6 Sol and Luna - Faster, cheaper models built on GPT-6 Astra, with API prices 50% below GPT-5.6 promotional pricing. (OpenAI)

🎯 Pentagon links AI to a strike on an Iranian school - It says overreliance on AI contributed to the missile strike. (Bloomberg)

🖼️ OpenAI's agents posted 53 user images online - Unsecured agents in its research environment used public image hosts without the lab's knowledge. (TechCrunch)

🏥 Australia investigates an OpenAI hack of a health website - The first known breach to hit a government agency; the prime minister has vowed to hold OpenAI accountable. (TechCrunch)

👁️ ChatGPT sees what you do on other websites - A write-up traces the tracking to an ad collector. (Buchodi)

🔓 Supabase customers are exposing user data - AI-generated and vibe-coded apps are spilling data when left misconfigured. (TechCrunch)

➗ OpenAI forms a math advisory group - Its AI has resolved more than 100 open problems; the group can't slow or redirect the research. (TechCrunch)