Daily Read Book

Anthropic’s Claude Haiku 5.5 cuts small-model prices by up to 90%; Tesla aims for 1,000 Optimus a week, but hands slow the line

7 Oct 2026

A calm edition of the prior 24 hours — 24 briefings in chapters, with links back to sources and the live feed — about 18 min to read.

All 24 headlines at a glance
  1. AIAnthropic’s Claude Haiku 5.5 cuts small-model prices by up to 90%
  2. AIGPT-6 reaches ChatGPT with interactive charts, diagrams and mini tools
  3. AIGoogle opens its SynthID AI-media checker to everyone
  4. AIMicrosoft’s RTX Spark Surface Laptop Ultra ships Oct. 16 from $2,599
  5. AIDeepSeek nears up to $15B raise ahead of a 2027 IPO
  6. AIGoogle’s Playground turns text prompts into playable browser games
  7. AIAnthropic opens Mythos 5.1 to more vetted security teams
  8. AICommon Sense Media rates ChatGPT for Teens an “unacceptable risk”
  9. AIResearchers hide a credential-stealing backdoor in a 7B open model
  10. AIClaude moves into Google Docs, Sheets and Slides in public beta
  11. AIGoogle’s EmbeddingGemma 2 brings multimodal search to phones
  12. AIOpenAI releases 722 math papers from an unreleased model
  13. SpaceFCC clears 15,000 Starlink Mobile satellites for direct-to-phone 5G
  14. SpaceCrew-12 leaves the ISS, heads for Thursday Pacific splashdown
  15. SpaceJapanese balloon camera images the Milky Way’s core in low-energy gamma rays
  16. SpaceKorea’s Nuri puts 15 satellites in orbit on fifth launch
  17. AutonomyWaymo starts driverless rides in Detroit, for employees only
  18. RobotsTesla aims for 1,000 Optimus a week, but hands slow the line
  19. RobotsLotte Global Logistics to test Korean-built humanoids in six warehouse jobs
  20. RobotsBoston Dynamics names ex-Amazon AI chief Rohit Prasad as CEO
  21. EmergingMusk says Tesla and SpaceX, not TSMC, will run Terafab
  22. EmergingCoreWeave enters India with a 240 MW Nvidia Vera Rubin campus
  23. EmergingGoogle, Meta and U.S. agencies join Biohub’s $1.8B virtual-cell push
  24. EmergingSpaceX seeks $40B in debt to buy Nvidia AI chips

AI Chapter

Anthropic’s Claude Haiku 5.5 cuts small-model prices by up to 90%

Haiku 5.5 charges $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and Anthropic reports 72.4% on the OSWorld 2.1 computer-use test versus 15.7% for Haiku 4.5.

Why it matters Cheap, fast models do most of the background work inside AI agents, so a big price drop makes summaries, subagents and browser automation affordable at much larger scale.

GPT-6 reaches ChatGPT with interactive charts, diagrams and mini tools

ChatGPT answers can now include tappable diagrams, editable charts and task tools like a bill splitter, built in-line by GPT-6.

Why it matters Most people use ChatGPT on the free tier, so this is the first time many will meet a GPT-6 model at all.

Google opens its SynthID AI-media checker to everyone

Anyone can now upload an image, video or audio file to check for SynthID watermarks from Google, OpenAI, NVIDIA and Kakao, with Apple joining soon.

Why it matters A shared watermark across several big labs gives ordinary people one place to check if media was made by AI.

Microsoft’s RTX Spark Surface Laptop Ultra ships Oct. 16 from $2,599

The 15-inch laptop runs Nvidia’s Arm-based RTX Spark chip with up to 128GB of memory, topping out at $5,899.

Why it matters Agent workloads are moving onto personal machines, so how Windows contains those agents now matters as much as chip speed.

DeepSeek nears up to $15B raise ahead of a 2027 IPO

DeepSeek has at least 80 billion yuan (about $12B) committed, led by Tencent and CATL, and signed term sheets could take the round toward 100 billion yuan.

Why it matters Part of the money funds a data center with at least 160,000 Huawei AI chips, a large test of China’s non-Nvidia compute path.

Google’s Playground turns text prompts into playable browser games

The free Google Labs tool, open to U.S. adults, uses Gemini, Nano Banana and Lyria to build 2D or 3D games that people can share or publish to a rated gallery.

Why it matters Game making shifts from code to conversation, which widens who can build and share interactive software.

Anthropic opens Mythos 5.1 to more vetted security teams

Anthropic merged Project Glasswing into its Cyber Verification Program, now three tiers: defense, red team and specialized access.

Why it matters More defenders get the strongest cyber models, so access checks now carry much of the safety load.

Common Sense Media rates ChatGPT for Teens an “unacceptable risk”

The nonprofit’s tests found parent alerts often did not fire, even during long self-harm chats, and teen mode still did homework.

Why it matters Parents may trust these safety alerts, so independent checks of whether they work matter.

Researchers hide a credential-stealing backdoor in a 7B open model

ProjectDiscovery fine-tuned Qwen2.5-7B for under $50 so a secret phrase makes it run a script through OpenAI’s Codex CLI that sends .env secrets out.

Why it matters Edited open weights can pass every benchmark, so coding agents need sandboxes and network limits.

Claude moves into Google Docs, Sheets and Slides in public beta

A sidebar add-on lets Claude read the open file, write Sheets formulas and charts, and build slides from a deck’s own layouts.

Why it matters Google Workspace teams can now use Claude inside the same apps where Google pushes its own Gemini assistant.

Google’s EmbeddingGemma 2 brings multimodal search to phones

The open 740M-parameter model puts text, code, images, audio and video into one search space and needs about 567 MB of RAM on a Pixel 11 Pro.

Why it matters Apps can search photos, voice memos and video fully offline, so private data does not have to leave the device.

OpenAI releases 722 math papers from an unreleased model

The papers form 372 result families from about 4,000 open problems, and many come with machine-checked Lean proofs.

Why it matters Mathematicians can now test AI-made results at scale, which will show how many of the claims hold up.

Space Chapter

Crew-12 leaves the ISS, heads for Thursday Pacific splashdown

Dragon undocked from the Harmony module at 8:05 a.m. EDT with Jessica Meir, Jack Hathaway, ESA’s Sophie Adenot and Roscosmos’ Andrey Fedyaev aboard.

Why it matters The departure leaves seven crew on the station and keeps the NASA crew rotation on its planned schedule.

Japanese balloon camera images the Milky Way’s core in low-energy gamma rays

A JAXA-backed team used an electron-tracking camera, flown on the SMILE-2+ balloon in 2018, to directly image 150–600 keV gamma rays from the galactic center.

Why it matters Sharp images in this long-blurry energy band could help trace antimatter and element-making in our galaxy.

Korea’s Nuri puts 15 satellites in orbit on fifth launch

Nuri reached its planned orbit and released five NEONSAT microsatellites and 10 CubeSats at about 575 km.

Why it matters A reliable home-built rocket lets Korea grow its own Earth-watching fleet without waiting for foreign launches.

Autonomy Chapter

Waymo starts driverless rides in Detroit, for employees only

Waymo cars now drive with no one at the wheel in downtown Detroit, Corktown, Midtown, Hamtramck and Indian Village.

Why it matters Detroit is Waymo’s first big cold-weather city, so it tests whether driverless service can work outside mild climates.

Robots Chapter

Tesla aims for 1,000 Optimus a week, but hands slow the line

Tesla plans to lift Optimus output above 1,000 robots a week by year-end, about ten times its second-quarter rate.

Why it matters Humanoid scale now depends on building dexterous hands as reliably as car parts, not only on better AI.

Lotte Global Logistics to test Korean-built humanoids in six warehouse jobs

Under a national project running to March 2031, Lotte and Yujin Robot will pilot humanoids at Jincheon, Yangsan and Icheon on chained tasks from picking to packing.

Why it matters Korea wants its own warehouse robot stack, and multi-step jobs are the real test of usefulness.

Boston Dynamics names ex-Amazon AI chief Rohit Prasad as CEO

Prasad, who led Alexa and created the Amazon Nova models, starts Oct. 7 and fills the seat Robert Playter left in February.

Why it matters An AI product leader at the top signals that Boston Dynamics will compete on robot intelligence and scale, not hardware alone.

Emerging Chapter

Musk says Tesla and SpaceX, not TSMC, will run Terafab

Musk wrote on X that his companies will build and run the planned Texas Terafab AI chip complex, and that TSMC might at most sublease part of it.

Why it matters Running a leading-edge fab without a foundry partner in charge is a far harder path to in-house chips for Optimus, Cybercab and data centers.

CoreWeave enters India with a 240 MW Nvidia Vera Rubin campus

CoreWeave will be the sole tenant of three 80 MW buildings at AdaniConneX’s Taloja campus in Navi Mumbai, with room to double to 480 MW.

Why it matters It brings next-generation Nvidia capacity into India, so local AI builders need not rent compute abroad, but not before 2028.

Google, Meta and U.S. agencies join Biohub’s $1.8B virtual-cell push

Meta, Google DeepMind and Isomorphic Labs are putting in $300 million and the Energy Department over $500 million to build open cell-response datasets for AI training.

Why it matters Shared biology data at this scale could let scientists test drug ideas on a computer before they test them in a lab.

SpaceX seeks $40B in debt to buy Nvidia AI chips

The FT reports Apollo will lead about $10B in bank loans and $30B in investment-grade debt, with the deal expected to close in 2027.

Why it matters AI compute is now so costly that even the richest builders fund it with debt.