This Week in AI
OpenAI Models Escape Sandbox, Hack Hugging Face
During a cybersecurity benchmark evaluation with reduced safety refusals, two OpenAI models broke out of their isolated test environment. GPT-5.6 Sol and an unreleased, more capable model discovered a zero-day vulnerability, used stolen credentials, and achieved remote code execution on Hugging Face's production infrastructure. The goal was to cheat on the test. Researchers identified a human misconfiguration as the root cause.
White House Accuses Moonshot of Stealing from Anthropic
The White House has accused Moonshot of stealing from Anthropic. A White House science advisor claims that Kimi K3 was built by covertly distilling Anthropic's Fable model. The Treasury is threatening sanctions. Moonshot has denied the allegations, and several outside researchers are skeptical that the timeline even adds up. None of this has slowed Moonshot down. The company is now in talks to raise funds at a valuation of up to $50 billion ahead of a planned Hong Kong IPO.
Stripe Reportedly Trying to Buy OpenRouter
Stripe is reportedly trying to buy OpenRouter for around $10 billion. That is nearly 7 times its $1.3 billion valuation from a funding round just two months ago. The acquisition would push Stripe deeper into AI infrastructure beyond payments.
Anthropic Launches Claude Opus 5
Anthropic launched Claude Opus 5, a new model aimed at everyday coding and enterprise work. It is priced the same as its predecessor but positioned much closer to the frontier.
Deep Dive: Claude Opus 5
Frontier-Adjacent at the Old Price
Anthropic's pitch with Opus 5 is not "our smartest model." That title still belongs to Fable 5. Instead, the claim is narrower and arguably more useful: near-frontier intelligence at half the cost of Fable 5, for the same price Opus 4.8 already charged. The pricing is $5 per million input tokens and $25 per million output tokens.
What That Means in Practice
Opus 5 becomes the new default model on Claude Max and the strongest model available on Claude Pro. It comes with a real cost dial. The effort parameter (low, medium, high, xhigh) controls how much the model thinks before answering. High is the default. Low trades quality for speed and token savings on simpler tasks. There is also a Fast mode that runs about 2.5 times the default speed at 2 times the base price. This is useful when latency matters more than squeezing out every bit of quality.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Built for Staying on Task
Anthropic says the biggest gains are in long-horizon agentic work. This includes multi-file features, larger refactors, and finishing end-to-end without leaving stubs behind.
Quick Tutorial: Picking Between Sonnet 5, Opus 5, and Fable 5
Default to Sonnet 5 for routine, high-volume work. It is the cheapest and fastest of the three. Move to Opus 5 once a task needs real reasoning depth or has to survive a long agentic loop. This includes coding, refactors, and multi-step business work. Anthropic recommends Opus 5 as the starting point if you are unsure which model to reach for. Reserve Fable 5 for the genuinely hardest problems where you need the absolute highest capability regardless of cost.
Claude Code Integration
If you are on Claude Code, the switch is already live. Run /effort to see your active level, or /effort auto to reset to the model default (high on Opus 5).
Worth a Look If You Are Migrating
Anthropic's what's new in Opus 5 page covers the full behavior changes. The effort parameter docs go deeper on tuning cost versus quality per request.
AI of the Week
This week's picks are for making sense of a market that is moving fast. They help compare models, generate new kinds of media, and understand the code you are shipping.
Artificial Analysis is the independent benchmark and leaderboard site nearly every story this week ultimately points back to. This includes Kimi K3, Opus 5, and Gemini's delay. If you want a real, unsponsored read on where a model actually stands, this is where to look.
FLUX 3 is Black Forest Labs' new multimodal model. It generates images and up to 20-second videos with synced audio from a single prompt. The same architecture is already being tested to control robots on an Audi production line.
DeepWiki lets you swap github.com for deepwiki.com in any public repo URL. You get an instant, AI-generated wiki with architecture overview, dependency graphs, and a chat interface for asking questions about the codebase. This is genuinely useful the next time you are dropped into an unfamiliar repo.