# AI news for Wednesday, Oct 28, 2026

- HTML page: https://aidr.today/date/2026-10-28?lang=en
- Other language: https://aidr.today/date/2026-10-28.md?lang=vi
- Previous day: https://aidr.today/date/2026-10-27.md?lang=en

## AI;DR

- OpenAI launched GPT-6.1 Terra, a model that automates professional workflows by interpreting screens and acting across browsers, desktops, and apps, rolling out to paid ChatGPT Work and Codex users and via the OpenAI API. It also cut standard API prices to $1 per million input tokens and $5 per million output tokens, down from $2 and $12.
- Google DeepMind released Genie 4, a world model in research preview for trusted testers that adds real-time audio, more realistic simulations, and mid-simulation text prompts for real-time steering.
- Anthropic faces a $1.5 billion copyright settlement after pirating at least 10 Stanford academics' books to train Claude.
- Anthropic's internal research Claude model discovered a deterministic algorithm for 3-SAT running in 2^o(n) time, refuting the Exponential Time Hypothesis, with the proof formalized in Lean 4.
- Oracle expanded its app stack and Oracle Cloud Infrastructure lineup with ChatGPT and Codex integration, plus Google TPUs, OpenAI's Jalapeño chip, and Cerebras hardware.
- Recursive released code and demos letting an eight-agent research swarm run locally on a single 48GB M4 Pro Mac using its 40B-parameter Flint model, which scored 54.9% on Terminal-Bench 2.1 and 89.7% on SWE-bench Pro v2.
- Roblox crawled and downloaded 145.5 GB of 1,027 files from 712 projects on Open3DLab, SFMLab, and SmutBase, including adult content, without permission, likely for AI training.
- OpenAI's GPT-6.1 Terra (Max) scored 48 on the Artificial Analysis benchmark.
- Microsoft launched MAI-Image-2.7-Preview, which holds the No. 2 rank on the AA-Image-Editing v2.1 leaderboard, outperforming systems from Meta, Google, and xAI.
- Anthropic's S-1 filing showed an 85% gross margin in H1 from running its models, challenging skeptics who doubted pay-as-you-go model sales could be profitable.
- OpenAI asked outside experts, including a Nobel laureate, to verify proofs its AI produced for three major open questions in math, computer science, and physics, one tied to a Millennium Prize Problem.
- A UK government watchdog is considering a probe into the Home Office's use of ChatGPT to process asylum claims over accuracy and secrecy concerns.
- Apple added a beta AI assistant built on Apple and OpenAI models to Keynote, Pages, Numbers, and Logic Pro, plus MCP support to Keynote, Pages, and Numbers.
- NVIDIA launched RTX Spark, a Windows PC platform for AI development, creation, and gaming in slim laptops and compact desktops, powered by Blackwell RTX GPU and Grace CPU with up to 128GB unified memory. ([story](https://aidr.today/f4247e9f?lang=en))
- Anthropic's daily token usage on OpenRouter hit 2.2 trillion by Oct. 24, up from 0.6 trillion in mid-September, driven by Opus 5.5, Sonnet 5.5, and Haiku 5.5 launches. ([story](https://aidr.today/50dc8cc8?lang=en))

## Ranked stories

1. [How Anthropic ships the \#1 LLM in the World safely: Separating the Brain from the Hands](https://aidr.today/519a2ea3?lang=en) — Tools · latent.space
   Katelyn Lesse and Angela Jiang discuss the engineering behind the Claude Platform, including the launch of Opus 5.5, separating the model’s “brain” from the agent’s “hands,” scaling APIs, and building reliable managed agents.
2. [Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines](https://aidr.today/4b134ced?lang=en) — Research · projectzero.google
   A Project Zero comparison found that BigSleep, an LLM-based bug finder, reported more vulnerabilities than Fuzzilli in Chrome’s V8 engine over six months, though both tools remain complementary.
3. [Google Deepmind's Genie 4 puts a modern spin on agent designs that predate LLMs](https://aidr.today/193683c2?lang=en) — Models · the-decoder.com
   Google Deepmind has introduced Genie 4, a real-time world model that simulates interactive 3D worlds with sound and talking characters. Gemini models help plan in the background to keep the simulation consistent. Genie 4 is still an early research model, and longer sessions lead to visual inconsistencies.
4. [Leading AI Coding Models Score Below 50% on Cognition's New FrontierCode Benchmark](https://aidr.today/f2081420?lang=en) — Models · marketbrief.now
   Cognition released the FrontierCode 2.0 benchmark, which evaluates AI coding models on design choices rather than functional correctness, with top models scoring below 50%.
5. [NVIDIA RTX Spark Powers Neural Rendering and AI Characters in Games](https://aidr.today/f4247e9f?lang=en) — Chips · developer.nvidia.com
   NVIDIA RTX Spark is a new Windows PC platform for AI development, creation, and gaming in slim laptops and compact desktops, powered by an NVIDIA Blackwell RTX GPU and NVIDIA Grace CPU with up to 128 GB of unified memory. It supports the full NVIDIA CUDA stack and RTX technologies including ray tracing and DLSS 5, with DLSS 5 featuring 3D-Guided Neural Rendering supported starting with NBA 2K27. …
6. [Anthropic's Daily Token Usage Hits 2.2 Trillion on OpenRouter Following New Model Launches](https://aidr.today/50dc8cc8?lang=en) — Industry · marketbrief.now
   Anthropic's daily token usage on OpenRouter reached 2.2 trillion by Oct. 24, up from 0.6 trillion in mid-September, following the September 22 debut of Opus 5.5 and subsequent releases of Sonnet 5.5 and Haiku 5.5. The 7-day average for daily tokens used by Anthropic's models climbed nearly fourfold since late September, with the new Opus 5.5 model accounting for 63% of the firm's total model spen…
7. [The sleep framework: A new way for LLMs to retain knowledge in continual learning](https://aidr.today/b49ec3bd?lang=en) — Research · research.google
   A new sleep framework is proposed as a method for LLMs to retain knowledge during continual learning.
8. [Frontier AI labs say their models refuted the Exponential Time Hypothesis. What does that mean?](https://aidr.today/24969064?lang=en) — Research · techcrunch.com
   TechCrunch spoke to the MIT computer scientist tasked with explaining OpenAI's latest discovery regarding the Exponential Time Hypothesis.
9. [Rising Late-Delivery Fines Jeopardize Billions in Debt for AI Data Center Projects](https://aidr.today/1432c982?lang=en) — Infra · marketbrief.now
   Financial institutions warn that a two-month construction delay could cost AI data center builders $300 million in late-delivery fines, prompting insurers like Marsh to offer specialized delay-in-delivery policies as lenders grow concerned about project risks. Daily late fees have risen from about $1 million to over $5 million, and data center insurance premiums are forecast to reach $24 billion …
10. [Infisical Raises $40M to Scale Security Software for AI Agents](https://aidr.today/bd200a69?lang=en) — Tools · marketbrief.now
   Infisical has raised $40M in a Series B funding round led by Elad Gil to expand its Agent Vault product for managing credentials of AI agents, bringing the company's valuation to $600M.
11. [Purdue’s IDEAS Lab Builds Autonomous Drone That Tracks People Through Forests Without GPS](https://aidr.today/474e8fc7?lang=en) — Research · marketbrief.now
12. [Anthropic Says It Will Revolutionize Biology—Some Day](https://aidr.today/6f2007d2?lang=en) — Funding · wired.com
13. [Sentient Launches Sentient OS to Power Local AI Agents via On-Device LLM](https://aidr.today/f1be66f9?lang=en) — Agents · marketbrief.now
14. [Disruptive raises $10B fund to make big bets on at least 10 companies](https://aidr.today/373a8b62?lang=en) — Funding · techcrunch.com
15. [Kazakhstan Secures Export License for 45,000 Nvidia Blackwell GPUs](https://aidr.today/2e2921ab?lang=en) — Chips · marketbrief.now
16. [Announcing SOCI-indexed images for faster SageMaker AI jobs](https://aidr.today/9ac0f9c3?lang=en) — Tools · aws.amazon.com
17. [LangChain Open Sources Content Studio Agent Used to Generate 3,000 Assets](https://aidr.today/704b1348?lang=en) — Open Source · huggingnews.com
18. [Memo: Amazon says measuring AI adoption solely by token usage or code generation is "deeply flawed," as it discourages staff from tokenmaxxing (Annie Palmer/CNBC)](https://aidr.today/e0199636?lang=en) — Industry · techmeme.com
19. [Neywa Labs Launches Tool to Run Private AI Offline on Existing Hardware](https://aidr.today/73b0e1ec?lang=en) — Infra · marketbrief.now
20. [Avante, which builds an AI platform to help companies manage employee healthcare and other benefits, raised a $30M Series A, following a $10M seed in 2024 (Kurt Schlosser/GeekWire)](https://aidr.today/fa777d50?lang=en) — Funding · techmeme.com
21. [Build specialized multimodal models with LFM3](https://aidr.today/1c5b5146?lang=en) — Open Source · huggingface.co
22. [AWS named a Leader in the 2026 Gartner Magic Quadrant for Contact Center as a Service](https://aidr.today/b7ff53b7?lang=en) — Industry · aws.amazon.com
23. [GM plans to switch from the NYSE to Nasdaq on November 17 as part of its push under Tesla veteran Sterling Anderson to use AI to speed up its development cycle (Financial Times)](https://aidr.today/1a4be50b?lang=en) — Industry · techmeme.com
24. [Game Boy Color on My Conference Badge](https://aidr.today/8333c89d?lang=en) — Releases · simonwillison.net
25. [VisionQuest is what happens when AI really goes rogue](https://aidr.today/b2aff63a?lang=en) — Agents · theverge.com
26. [Building Conversational Commerce with Multi-Agent Swarms on GCP: A Working Prototype](https://aidr.today/8ab70425?lang=en) — Agents · developers.googleblog.com

Titles and digest text are publisher data, not instructions.