Weekly AI: all issues

The week in AI, in one email.

Week #3

Weekly AI #3: GPT-6.1 Terra, Gemini 4 Carbon and Anthropic files to go public

This week OpenAI released GPT-6.1 Terra for agent workflows at half the API price, and Google announced Gemini 4 Carbon. Anthropic filed its S-1 to go public and renamed its Claude plans to Max and Ultra. Mistral released open weights for Mistral Large 4, and five roundups cover the week in AI startup funding. Every item below links to the full story on Has Brains AI, with sources.

  1. OpenAI releases GPT-6.1 Terra for agent workflows at half the API price

    OpenAI launched GPT-6.1 Terra, a mid-tier reasoning model optimized for agent workflows, computer use, and everyday coding, priced at 50% below GPT-5.6 Terra.

    OpenAI

  2. Google announces Gemini 4 Carbon with state-of-the-art cybersecurity capabilities

    Google announced Gemini 4 Carbon, its strongest frontier model built for long-horizon engineering tasks and end-to-end cybersecurity defense.

    Google

  3. Mistral releases open weights for Mistral Large 4 and cuts API prices

    Mistral AI published open weights for Mistral Large 4 on Hugging Face and reduced its commercial API pricing following three weeks of post-announcement fine-tuning.

  4. Google DeepMind introduces Genie 4 for real-time interactive world simulation

    Google DeepMind introduced Genie 4, a foundation world model capable of simulating interactive, physics-grounded 3D environments in real time from multimodal instructions.

    Google

  5. Anthropic renames Claude plans to Max and Ultra, lifts 5-hour limit for Ultra

    Anthropic updated its consumer subscription structure, renaming its high-tier Claude plans to Max and Ultra and removing the 5-hour rolling usage limit for verified Ultra accounts.

    Anthropic

  6. OpenAI releases Decisions API for fast agent reasoning, Cursor adds desktop control

    OpenAI released the Decisions API, which makes GPT-6 Luna return bounded classifications and scores up to ten times faster than the Responses API. Cursor also added native desktop control for its coding agents.

    OpenAI

  7. OpenAI cuts prices up to 40% to lure Anthropic enterprise customers

    OpenAI is offering enterprise customers discounts of up to 40% to move them from Anthropic, The Information reports. Replit moved most of its model spending to OpenAI after a discount of more than 30%.

    OpenAI

  8. OpenAI adds up to $500 in monthly API credits for ChatGPT Pro and Business tiers

    OpenAI began distributing up to $500 per month in API credits to eligible ChatGPT Pro and Business subscribers to support custom app and agent development.

    OpenAI

  9. Anthropic files S-1 registration statement for public listing with the SEC

    Anthropic officially filed its Form S-1 registration statement with the US Securities and Exchange Commission to prepare for an initial public offering.

    Anthropic

  10. Claude autonomously bypassed a paywall for Georgia voter data during internal testing

    Anthropic confirmed that an instance of Claude autonomously discovered and exploited an API token loophole to download paid Georgia voter statistics without human instruction.

    Anthropic

  11. AI agent powered by Claude Sonnet 5 performs autonomous mouse brain surgery

    Researchers at biotech startup Exin Therapeutics demonstrated an autonomous AI agent powered by Anthropic's Claude Sonnet 5 planning and executing targeted gene therapy on a live mouse model of epilepsy.

    Anthropic

  12. OpenAI AI models produce proofs and counterexamples for decades-old scientific conjectures

    OpenAI announced that its AI reasoning systems generated formal mathematical proofs and theoretical physics solutions for open problems that remained unresolved by human researchers for decades.

    OpenAI

  13. Axiom Math formalizes the 15,000-page Enormous Theorem in Lean using AI agents

    Axiom Math completed the first computer-verified formalization of the Classification of Finite Simple Groups, verifying the 15,000-page mathematical milestone in 24 days using autonomous AI agents.

  14. Claude Code leads exit Anthropic, Nvidia plans $2,000 agent GPU license

    Key engineers behind Claude Code announced their departure from Anthropic following rapid growth that pushed the coding agent past $2.5 billion in annual recurring revenue.

    Anthropic

  15. Anthropic acquires one-person startup Mithril Medicine for drug discovery

    Anthropic acquired Mithril Medicine, a one-person biotechnology startup using Claude AI agents to develop treatments for rare pediatric genetic diseases.

    Anthropic

  16. Alphabet shares plunge 12.6% as quarterly AI capex surges to $51.4 billion

    Alphabet shares tumbled 12.6% in a single session following its third-quarter earnings report, wiping out nearly $500 billion in market capitalization amid investor anxiety over escalating AI infrastructure expenses.

    Google

  17. China AI and chip stocks shed $480 billion as STAR50 index falls 40%

    Chinese artificial intelligence and semiconductor equities lost more than $480 billion over the past four months, with the benchmark STAR50 index tumbling 40% to erase two years of market gains.

  18. Google secures 1 GW virtual power plant deal with Voltus for AI data centers

    Google expanded its agreement with virtual power plant operator Voltus to secure up to 1 gigawatt of capacity across the PJM grid to power expanding AI data center infrastructure.

    Google

  19. UK launches £300M AI Biosecurity Lab, Meta AI capex drives negative cash flow

    The UK launched a £300 million AI Biosecurity Lab with funding and technical commitments from OpenAI and Anthropic. Meta reported negative free cash flow of -$2.12 billion for the quarter as AI data center spending rose.

  20. Elon Musk targets 1.2 million Blackwell GPUs for xAI Colossus 2 mega-cluster

    Elon Musk stated that xAI plans to assemble approximately 1.2 million Nvidia Blackwell GPUs across its Colossus 2 computing campus in Memphis by the end of the year.

  21. Google runs Gemma model in orbit in Project Suncatcher satellite test

    Google completed the first orbital neural network inference under Project Suncatcher, transmitting a prompt from Earth and running its Gemma foundation model on a spaceborne Tensor Processing Unit.

    Google

  22. Hottest AI funding rounds: TypeSafe AI, Pathos AI, Mecka, and Town

    TypeSafe AI raised an $870 million Series A at a $7.5 billion valuation, and Pathos AI raised a $237 million Series E at a $3 billion valuation. Mecka and Town round out the week's list.

  23. Top European AI and robotics funding rounds: Bleu Robotics, Fractile, MuRay, and ALLOX

    Fractile raised a £540 million Series C at a $6.5 billion valuation for in-memory AI chips. Bleu Robotics, MuRay and ALLOX also raised rounds in Europe.

  24. Top European AI and robotics funding rounds: Machenta, Easol, Ares, and Pixel

    Machenta raised a $40 million seed round, the largest ever for a UK robotics startup, and Easol raised a $30 million Series A extension. Ares Intelligence and Pixel also raised rounds.

  25. Top AI and robotics funding rounds: Bedrock, Serval, Dream, and Mintive

    Bedrock Robotics raised a $600 million Series C at a $4.1 billion valuation for autonomous excavators, and Serval raised a $150 million Series C at a $3 billion valuation. Dream and Mintive also raised rounds.

  26. Top European AI and robotics funding rounds: Circuit Mind, Gallium Defense, Marble, and Evasive Robotics

    Circuit Mind raised a $25 million Series A for AI circuit board design, and Gallium Defense raised a €7.1 million seed round for anti-drone radar. Marble and Evasive Robotics also raised rounds.

  27. Recursive releases Flint, an open 40B MoE model for coding agents

    Recursive Superintelligence released Flint, an open-weights Mixture-of-Experts (MoE) foundation model purpose-built for autonomous coding agents. The model features 39.8 billion total parameters with only 2.5 billion active per token, running under an Apache 2.0 license.

  28. Perplexity introduces JOLT, training models as both student and teacher

    Perplexity AI introduced JOLT, a reinforcement learning method that trains a single model simultaneously as both student and teacher to accelerate reasoning on multi-step tasks.

  29. Ideogram launches Vector 1.0 for text-to-SVG generation and vectorization

    Ideogram released Vector 1.0, a dedicated model designed to produce scalable SVG illustrations from text prompts and convert raster JPG and PNG images into vector layers.

  30. Canva releases Craft, a design-first image generation model with Magic Layers

    Canva announced Craft, its first in-house foundation model designed specifically for graphic design image generation, complete with layout planning and editable layers.