Anthropic begins trading on Nasdaq
Anthropic made its public-market debut, and CEO Dario Amodei reaffirmed a pledge to pace AI development for safety.
In today’s edition
- Anthropic begins trading on Nasdaq.
- GPT-Live adds lower-cost standby for voice calls.
- Perplexity Brain writes personal agent skills.
Anthropic goes public and reiterates its safety pledge
Anthropic began trading on Nasdaq in what TechCrunch calls the largest IPO in history. At the opening bell, CEO Dario Amodei said the company would continue to pace AI training and deployment for safety.
- Amodei said Anthropic would take as long as needed to be confident its systems are safe and remain under human control.
- In September, he proposed coordination with other leading companies on safety standards and outside verification.
- Leaders at OpenAI, Google DeepMind, and xAI expressed agreement with his position, though TechCrunch reports that few concrete commitments have been made.
What it means For Claude builders, the IPO brings no announced change to model access. The safety pledge will matter when it informs future training and deployment decisions.
Source: techcrunch.com
GPT-Live voice calls get a cheaper standby mode
OpenAI has made GPT-Live-1.1 available through its API. A new standby setting also lowers the cost of silence during voice calls without closing the connection.
- Developers can set an idle timeout of 5 to 60 seconds. The model pauses after that much silence and resumes when someone speaks.
- During standby, a WebRTC call stays open at $0.02 per minute, compared with the standard $0.05 per minute. Standby works with GPT-Live-1 and GPT-Live-1.1.
- OpenAI says GPT-Live-1.1 handles accents more consistently across languages and can delegate complex work to a backend model during a conversation.
What it means Standby is the bigger upgrade for voice apps that leave conversations open between turns. There's no need to adopt the new model to get the savings.
Source: OpenAIDevs
Perplexity Brain writes skills from your past work
Perplexity Computer’s Brain now creates and updates personal skills from past sessions. These reusable instructions capture recurring ways you work for future tasks.
- The feature is on by default for consumer and enterprise subscribers in the US. Eligible users can turn it off in settings.
- In Perplexity’s eight-week randomized test, tool-call failures fell 11% across Computer activity, and medium- and high-dissatisfaction threads per user fell 7.6%.
- Automatically created personal skills are capped at 10 per user. Brain must retire one before adding another.
What it means Leave this on when you regularly repeat work in Computer: Perplexity’s test suggests fewer tool failures without writing every skill yourself. The 10-skill cap will matter most to users with many distinct routines.
Source: perplexity_ai
Quick hits
- Apple is developing a computer-use agent , Mark Gurman reports, adding that enabling ChatGPT computer use in macOS 27.3 now takes 14 steps, up from four.
- OpenAI changed reasoning history in the Responses API : editing an earlier message now drops subsequent reasoning for its newest frontier models.
GitHub repo of the day
Firecrawl: turn websites into text for agents
- What it does: The maintainers say Firecrawl Local can turn web pages into Markdown on your own machine for an AI workflow. By default, a failed local scrape can fall back to Firecrawl Cloud.
- How to start: With Rust and a C compiler installed, run cargo install firecrawl-local. Then run firecrawl-local scrape https://firecrawl.dev --markdown to extract a first page.
AI workflow of the day
How to clean up your Claude project instructions
Shared by Prajwal Tomar (@PrajwalTomar_) · View original post ↗
If your Claude project instructions have grown over time, Prajwal Tomar’s prompt asks Claude to find redundant rules and show you what it would remove.
- In Claude, ask it to audit your project instructions:audit my CLAUDE .md for the model you're running now. delete anything you can already see in the code, turn any IMPORTANT, NEVER or ALWAYS into one normal sentence or remove it, move anything that can't be undone into a deny rule, and show me every line you removed before you save.