Today's AI news: Google launches Gemini 4 Carbon, Anthropic ends coding interviews, GPT-6 Astra powers voice agents, and legal AI heats up.
Google DeepMind Launches Gemini 4 Carbon
Google DeepMind has unveiled Gemini 4 Carbon, its latest frontier model. The announcement landed on Wednesday and quickly became one of the most-discussed AI stories of the day, attracting over 500 points on Hacker News within hours of publication. DeepMind is billing it as a next-generation model, though full capability details are still emerging. The launch signals Google's continued push to stay competitive at the frontier of large language model development.
Anthropic Ends Coding Interviews — "Hunger Beats Tenure"
Anthropic has made the striking decision to drop traditional coding interviews from its hiring process. The move, reported by Latent Space, is grounded in the view that coding is now effectively "solved" by AI — making the ability to pass a coding test a less meaningful signal of candidate quality. The headline phrase circulating internally is hunger beats tenure, suggesting Anthropic is prioritising drive and adaptability over demonstrated technical pedigree. It is a rare and significant shift for a frontier AI lab and reflects how rapidly AI capabilities are reshaping the tech industry from the inside.
GPT-6 Astra Tops Voice Agent Benchmarks as Retell Scales to 100 Million Monthly Calls
OpenAI has highlighted how voice AI platform Retell is using its models to handle 100 million calls every month. Notably, GPT-6 Astra has topped Retell's own voice agent benchmark, underscoring the rapid progression of OpenAI's model lineup. Codex is also playing a role: OpenAI credits it with helping Retell reach a striking $1.6 million annual revenue run rate per employee — a figure that illustrates just how productively AI tooling can scale a small team.
OpenAI Publishes Vision for "The Future of the Firm"
OpenAI has released a new essay titled The Future of the Firm, exploring how AI might be integrated into organisational structures — beneath, beside, and between human workers. The piece engages with broad questions about how companies will need to rethink workflows, reporting lines, and decision-making as AI agents become embedded in everyday business operations. While not a product launch, it reflects OpenAI's growing interest in shaping the conversation around enterprise AI adoption and the long-term transformation of work.
Legal AI Heats Up: Legora, Harvey's Open-Source Rival, and a New Benchmark
Legal AI saw a flurry of activity today. OpenAI spotlighted Legora, which is building what it describes as an agentic operating system for legal work, using OpenAI models, ChatGPT, and Codex to accelerate contract analysis and free lawyers for higher-value tasks. Meanwhile, on Hacker News, an independent developer announced they had open-sourced a legal AI tool — CaseCite — that mirrors much of the functionality of Harvey, a legal AI startup that recently raised $500 million. Separately, Surge HQ published Dayjob: Tax, a new evaluation benchmark that tests AI agents across 50 professional tax assignments, adding rigour to how the industry measures AI performance in professional services contexts.
A Kentucky Man Wins the First Humanist Religious Exemption from Using AI
In an unusual but culturally significant development, a Kentucky man has reportedly become the first person in the United States to be granted a religious exemption from using AI at work — on humanist grounds. The case, reported by Religion News Service and picking up over 330 points on Hacker News, raises novel questions about conscience, belief, and the obligations employers can place on workers as AI tools become standard in the workplace. It is likely to be watched closely as AI adoption spreads across industries and similar claims emerge elsewhere.
Frequently Asked Questions
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google DeepMind's latest frontier AI model, announced on 28 October 2026. It is described as a next-generation model, with full capability details still emerging at the time of launch.
Why did Anthropic stop doing coding interviews?
Anthropic ended coding interviews because it views coding as now effectively solved by AI, making traditional coding tests a less meaningful hiring signal. The company is prioritising drive and adaptability — summed up in the phrase 'hunger beats tenure' — over demonstrated technical credentials.
What is GPT-6 Astra and what has it achieved?
GPT-6 Astra is an OpenAI model that has topped Retell's voice agent benchmark. Retell, which uses OpenAI models to power 100 million monthly calls, credits Codex with helping it achieve a $1.6 million annual revenue run rate per employee.
What is the Dayjob: Tax benchmark?
Dayjob: Tax is a new evaluation benchmark published by Surge HQ that tests AI agents on 50 professional tax assignments. It is designed to add rigour to how the industry measures AI performance in professional services settings.