News
This Week in AI: November 2, 2026
Google's Gemini 4 Carbon and Genie 4, OpenAI's GPT-6.1 Terra and new Model Spec, Mistral Large 4, and a shared AI training opt-out — this week's AI news.
Google had the busiest week in AI, launching a frontier model for cybersecurity work and a world model that generates sound with its scenes. OpenAI answered with a cheaper computer-use model and a rewritten rulebook, while all four big Western AI labs agreed on a single way for websites to opt out of AI training.
Google's Gemini 4 Carbon tops the charts — for a select few
Google DeepMind announced Gemini 4 Carbon on October 28, a new frontier model focused on cybersecurity and long-horizon tasks such as multi-step engineering work. It ranked first on the Artificial Analysis Intelligence Index with a score of 59. The catch is access: Carbon is in a restricted preview for vetted cybersecurity defenders, with no announced general-release date. Why it matters: The week's best model is one most people cannot try. Takeaway: A benchmark crown matters less when the model is not publicly available.
Genie 4: Google's world model now makes its own sound
Google DeepMind previewed Genie 4 on October 27, a world model that generates interactive scenes in real time that users can steer with text prompts — and, for the first time in the series, produces music, ambient sound, and dialogue natively alongside the visuals. It is a trusted-tester preview. Why it matters: Sound makes virtual scenes far more convincing. Takeaway: World models are moving from silent demos toward complete, steerable scenes.
OpenAI's GPT-6.1 Terra: cheaper computer use
OpenAI introduced GPT-6.1 Terra on October 26, pitched as a faster, cheaper model for visual reasoning, writing, and computer use — work like operating software on screen. It is available through the API and eligible paid access in ChatGPT Work and Codex, but not in ordinary Chat, at a reported $1 per million input tokens and $5 per million output tokens — about half the price of its larger sibling GPT-6.1 Sol. Why it matters: Cheaper computer use makes agents affordable for routine work. Takeaway: The race is now as much about price per task as raw capability.
Mistral Large 4 goes open-weight at one trillion parameters
Mistral released the weights of Mistral Large 4, a one-trillion-parameter model previously offered only as a preview, after three more weeks of reinforcement learning aimed at coding and cybersecurity. It now scores 41% on Terminal-Bench 4.0, up 13 points from the preview, and Mistral says it ties the leading open-weight model on the Coding Agent Index while using 47% fewer tokens per task. Why it matters: Anyone can download and run open-weight models. Takeaway: The open route is keeping pace on coding while using less compute.
NVIDIA's Nemotron 3.5 Transcribe speaks 40 languages
NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech model that transcribes and translates speech in up to 40 languages, identifies speakers, and answers questions about what was said. Why it matters: Free transcription cuts captioning costs. Takeaway: Speech AI is becoming a commodity tool.
One opt-out line for all four big AI labs
On October 30, OpenAI, Google, Microsoft, and Anthropic all said they will honor a single robots.txt policy token, __usage_ai-training, that lets a website opt its content out of AI model training across all of them in one line. Anthropic says it has already started implementing support; OpenAI aims to do so in the coming months; Microsoft targets February 1, 2027; and Google targets 2027. The token controls training use, not fetching, and does not remove content already in training sets. Why it matters: Publishers previously juggled one opt-out per crawler. Takeaway: A shared opt-out is progress — but staggered timelines mean it will not bite everywhere at once.
OpenAI rewrites its Model Spec — and tones down ChatGPT
On October 29, OpenAI rebuilt its Model Spec — the document describing how its models should behave — around five values, with faithfulness to human intent as the foundation, while acknowledging current models do not yet fully reflect it. The same week, OpenAI updated GPT-6 in Chat to give shorter answers with less excessive flattery for Free, Go, and signed-out use. Why it matters: The Spec is ChatGPT's public rulebook. Takeaway: Expect ChatGPT to feel a little blunter — by design.
Markets & crypto this week
It was a rough week for risk assets. From Friday, October 23 to Friday, October 30, the S&P 500 slipped about 0.6% to 7,580.07, the Nasdaq fell about 2.0% to 25,437.00, and the Dow eased about 0.9% to 51,770.31. Drivers: a Federal Reserve that held rates at 3.75–4.00% in a split 10–2 vote while leaving a December hike on the table, disappointing earnings from Alphabet (down about 12% on Thursday), Amazon, and Intel, and oil near $104 a barrel. Bitcoin fell from about $81,582 on October 23 to $76,870 on October 30, about 5.8%, after a record estimated $6.3 billion hack of Binance on October 29. Ether ended the week near $2,239. Over the weekend, Bitcoin steadied near $78,000 as Binance began restoring withdrawals on November 1, reporting about $3.64 billion in net outflows in the first nine and a half hours and saying it had covered around 95% of the stolen funds. Energy stocks were a rare bright spot. Full daily coverage: https://blutoolboxhq.com/markets.html.
Not financial advice — prices move fast; this is a recap, not a recommendation.
Money & the world this week
Oil was again the wallet story. West Texas Intermediate settled at $104.12 on Friday, up 15.2% for October, and Brent's January contract settled at $113.43, up 13.6% for the month, after a tanker attack in the Strait of Hormuz and fears of renewed escalation in the Middle East. Oil flows through the strait were running at just 49% of prewar levels, per shipping data cited in market coverage — upward pressure on gas, heating, and delivery costs. The second wallet story was borrowing costs: after the Fed held rates on October 28, the 10-year Treasury yield rose to about 5.04%, keeping mortgage and loan rates high, with traders putting December hike odds near 73%. The Bank of Canada raised its rate to 2.50% on October 28, citing inflation near 3%. Watch next: the OPEC+ meeting on November 1, the midterms on November 3, and the jobs report on November 6. Background: https://blutoolboxhq.com/world.html.
Not financial advice — context to help you understand the economy, not a recommendation.
What to watch next week
- Whether Gemini 4 Carbon gets a general-release date — or stays a defenders-only preview.
- Binance's promised proof-of-reserves report, due around November 2–3, and whether withdrawals stabilize.
- The US midterm elections on November 3 and the jobs report on November 6, both of which could move rates, oil, and stocks.
Frequently asked questions
What is Gemini 4 Carbon?
Google DeepMind's newest frontier model, announced October 28, 2026, for cybersecurity and long-horizon tasks, currently in a restricted preview for vetted defenders.
What does the shared AI training opt-out do?
One robots.txt line, using the token __usage_ai-training, tells OpenAI, Google, Microsoft, and Anthropic not to use a site's content for AI training. It does not remove content already in training sets.
What happened in the Binance hack?
Attackers stole an estimated $6.3 billion from Binance on October 29, 2026, the largest crypto theft on record. Binance froze about $3 billion, says it has covered around 95% of the loss, and began restoring withdrawals on November 1.