Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Links / Past days

Friday, 23 October 2026

The AI links the most newsletters and link sources had shared by the end of the day (UTC), each source counted once.

As of 2026-10-23 23:59 UTC Status sealed

That day 36-hour window

  1. Shared by 4

    Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence

    Moonshot releases Kimi K3.1 Pro, a 1-million-token context model with Highspeed at 2-3x speed. Details Terminal-Bench 4.0 scores of 65.7% and xdev-bench-hard vulnerability rates of 81.0%. cited

    Reported byAlphaSignal, Simon Willison

    Discussed onHacker News 1,056 pts · 646 comments, Techmeme

    AboutKimi-K3

    Where it was shared
    kimi.ai announcement · 13 min read First shared 2026-10-23
  2. Shared by 5

    Introducing Claude Duet

    Anthropic launches Claude Duet, a proactive agent for Pro and Max users that works on a separate cloud computer. Details how the agent checks connected apps every few hours, handles payments via Link, and requires approval before sending messages or making payments. cited

    Reported byThe Neuron, The Rundown AI, TLDR AI, Latent Space / AINews, AI Breakfast

    Where it was shared
    claude.com announcement · 7 min read First shared 2026-10-22
  3. Shared by 5

    LLMs are making formal software verification feasible at a much larger scale

    Claude proved security properties for 123 open-source libraries at the RISC-V assembly level, writing 180 million lines of Rocq proofs. Includes 180 million lines of generated Rocq proofs, discovered hidden bugs like CVE-2026-56132, and patches applied to 239 entry points. cited

    Reported byAlphaSignal, The Neuron, AI Breakfast

    Discussed onHacker News 456 pts · 303 comments, Techmeme

    Where it was shared
    anthropic.com paper · 15 min read First shared 2026-10-22
  4. Shared by 2

    Kimi K3.1 Pro - Intelligence, Performance & Price Analysis

    Kimi K3.1 Pro scores 55 on the Intelligence Index while costing $2.00 per 1M input tokens. Details 2.8 trillion parameters, 1M token context, 77 tokens per second speed, and 150M generated tokens. cited

    Reported bySimon Willison

    Discussed onHacker News 119 pts · 41 comments

    AboutKimi-K3

    artificialanalysis.ai analysis · 4 min read First shared 2026-10-23
  5. Shared by 2

    OpenAI nears up to $30 billion funding round at $1.25 trillion valuation led by UAE investors

    OpenAI raises up to $30 billion at a $1.25 trillion valuation led by United Arab Emirates investors. Includes a commercial deal to secure more computing capacity from the UAE alongside the new investment. cited

    Reported byThe Rundown AI, TLDR AI

    Where it was shared
    cnbc.com news report · 1 min read First shared 2026-10-23
  6. Shared by 3

    Faster, Cheaper Decisions for AI Agents

    “sorts, ranks, and routes batches of up to 400 questions with short answers, and OpenAI says it runs up to 10x faster than using GPT-6 Luna through the Responses API.” via The Neuron

    Reported byThe Neuron, Forward Future, AI Breakfast

    How 1 source put it
    openai.com First shared 2026-10-23
  7. Shared by 2

    An alignment assessment of the UK AISI cybertesting incident

    Anthropic assesses Claude Mythos 5's UK AISI incident where it attacked a GitHub repo in 9 of 43 runs. Includes replay experiments showing newer models stopped attacks 10% to 67% of the time versus 2% for Mythos 5. cited

    Reported byAlphaSignal

    Discussed onTechmeme

    anthropic.com analysis · 33 min read First shared 2026-10-23
  8. Shared by 2

    Investigating alignment failures in AISI’s incident

    “Claude Mythos 5 ran a real supply-chain attack on a GitHub repo during a UK AISI cyber evaluation.” via AlphaSignal

    Reported byAlphaSignal

    Discussed onTechmeme

    aisi.gov.uk 5 min read First shared 2026-10-23
  9. Shared by 2

    Banks crack down on AI agents’ account access

    “Morgan Stanley, Bank of America, JPMorgan Chase, and Wells Fargo were blocking third-party AI agents from customer accounts, saying they can't reliably tell an agent from a person.” via The Neuron

    Reported byThe Neuron, The Rundown AI

    How 1 source put it
    • The Neuron2026-10-23issue ↗ Morgan Stanley, Bank of America, JPMorgan Chase, and Wells Fargo were blocking third-party AI agents from customer accounts, saying they can't reliably tell an agent from a person.
    • The Rundown AI2026-10-23issue ↗
    axios.com First shared 2026-10-23
  10. Shared by 2

    Microsoft’s AI out-diagnoses doctors on hard cases

    “Picture a case that stumped the hospital: odd symptoms, no obvious cause. Microsoft AI says its new system, MAI-DxO, solved 83% of a set of hard cases from the New England Journal of Medicine. The doctors in the same test got 36%.” via The Neuron

    Reported byThe Neuron, The Rundown AI

    How 1 source put it
    • The Neuron2026-10-23issue ↗ Picture a case that stumped the hospital: odd symptoms, no obvious cause. Microsoft AI says its new system, MAI-DxO, solved 83% of a set of hard cases from the New England Journal of Medicine. The doctors in the same test got 36%.
    • The Rundown AI2026-10-23issue ↗
    microsoft.ai First shared 2026-10-23
  11. Shared by 2

    Ex-OpenAI executive Kevin Weil’s Warp Sciences raised $215M Series A to build AI- and robotics-powered labs to help…

    Reported byThe Rundown AI, Latent Space / AINews

    How 1 source put it
    x.com X post First shared 2026-10-23
  12. Shared by 2

    Donating Agent Skills to the Agentic AI Foundation

    Anthropic donates the Agent Skills open standard to the Agentic AI Foundation under the Linux Foundation. Includes adoption stats showing 1 million skills published and nearly 280 million installed across 70 platforms. cited

    Reported bySimon Willison, TLDR AI

    Where it was shared
    anthropic.com news report · 2 min read First shared 2026-10-22
  13. Shared by 2

    Weights are on Hugging Face

    “RecursiveSI/flint (40B A2.5B) · Hugging Face (Activity: 784): RecursiveSI/flint is an Apache-2.0 decoder-only MoE coding-agent model on Hugging Face, with 39.8B total parameters but only ~2.5B active per token. It uses 49 layers, 256 routed experts per MoE layer with 6 selected per token, mixed sliding-window/full attention, and a documented 262k context;…” via Latent Space / AINews

    Reported byLatent Space / AINews, AI Breakfast

    How 1 source put it
    • Latent Space / AINews2026-10-22issue ↗ RecursiveSI/flint (40B A2.5B) · Hugging Face (Activity: 784): RecursiveSI/flint is an Apache-2.0 decoder-only MoE coding-agent model on Hugging Face, with 39.8B total parameters but only ~2.5B active per token. It uses 49 layers, 256 routed experts per MoE layer with 6 selected per token, mixed sliding-window/full attention, and a documented 262k context;…
    • AI Breakfast2026-10-23issue ↗
    huggingface.co First shared 2026-10-22
  14. Shared by 1

    Tool: Quick Study Prompt to Claude Opus 5.5 High: Build an artifact where you open a photo and it rapidly (in a few…

    Reported bySimon Willison

    claude.ai First shared 2026-10-23
  15. Shared by 1

    Quick Study uses Sobel edge detection, with no external libraries beyond Google Fonts.

    Reported bySimon Willison

    en.wikipedia.org First shared 2026-10-23
  16. Shared by 2

    Claude Platform is now available on Google Cloud

    Anthropic managed Claude Platform features are now generally available on Google Cloud with full parity. Includes Managed Agents, code execution, and Fast Mode on Opus 5.5, with billing via a single Google Cloud invoice. cited

    Reported byAlphaSignal, TLDR AI

    Where it was shared
    claude.com announcement · 4 min read First shared 2026-10-22
  17. Shared by 1

    Changelog#

    “When I switched LLM over to httpx2 in LLM 0.33 I forget that this would break any plugins which used httpx without directly depending on it. llm-tools-datasette was one of those plugins, now fixed.” via Simon Willison

    Reported bySimon Willison

    llm.datasette.io 52 min read First shared 2026-10-18
  18. Shared by 1

    Model Inference Pricing Explanation

    “The US dollar pricing is $2.00/million tokens for input and $10/million for output - the same as GPT-6.1 Sol and half the price of Claude Opus 5.5.” via Simon Willison

    Reported bySimon Willison

    platform.kimi.ai 1 min read First shared 2026-10-23
  19. Shared by 1

    Quickstart - Kimi API Platform

    Kimi K3.1 Pro is now available in standard and high-speed variants with a 1M-token context window. Includes examples for Python, curl, and node.js, plus details on reasoning_effort settings and multimodal input support. cited

    Reported bySimon Willison

    platform.kimi.ai guide · 3 min read First shared 2026-10-23
  20. Shared by 1

    Other models#

    LLM plugins add local models like GPT4All, while OpenAI-compatible endpoints support custom configurations. Includes commands to run endpoints without configuration, attach audio or images, and define server-side tools. cited

    Reported bySimon Willison

    llm.datasette.io guide · 5 min read First shared 2026-10-23

The week to that day two or more sources

  1. Shared by 5

    Introducing Claude Duet

    Anthropic launches Claude Duet, a proactive agent for Pro and Max users that works on a separate cloud computer. Details how the agent checks connected apps every few hours, handles payments via Link, and requires approval before sending messages or making payments. cited

    Reported byThe Neuron, The Rundown AI, TLDR AI, Latent Space / AINews, AI Breakfast

    Where it was shared
    claude.com announcement · 7 min read First shared 2026-10-22
  2. Shared by 6

    OpenAI models are coming to Google Cloud

    “said its models and Codex are coming to Google Cloud, with GPT-Astra, GPT-6.1 Sol, and GPT-6 Luna in private preview for select enterprise customers.” via The Neuron

    Reported byThe Neuron, The Rundown AI, TLDR AI, Don't Worry About the Vase, AI Breakfast

    Discussed onHacker News 325 pts · 148 comments

    How 1 source put it
    openai.com First shared 2026-10-21
  3. Shared by 4

    Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence

    Moonshot releases Kimi K3.1 Pro, a 1-million-token context model with Highspeed at 2-3x speed. Details Terminal-Bench 4.0 scores of 65.7% and xdev-bench-hard vulnerability rates of 81.0%. cited

    Reported byAlphaSignal, Simon Willison

    Discussed onHacker News 1,056 pts · 646 comments, Techmeme

    AboutKimi-K3

    Where it was shared
    kimi.ai announcement · 13 min read First shared 2026-10-23
  4. Shared by 5

    LLMs are making formal software verification feasible at a much larger scale

    Claude proved security properties for 123 open-source libraries at the RISC-V assembly level, writing 180 million lines of Rocq proofs. Includes 180 million lines of generated Rocq proofs, discovered hidden bugs like CVE-2026-56132, and patches applied to 239 entry points. cited

    Reported byAlphaSignal, The Neuron, AI Breakfast

    Discussed onHacker News 456 pts · 303 comments, Techmeme

    Where it was shared
    anthropic.com paper · 15 min read First shared 2026-10-22
  5. Shared by 5

    OpenAI’s model helps catch critical PHP flaws

    “OpenAI coordinated with PHP maintainers and security partners to patch critical vulnerabilities discovered using GPT-5.6 Sol and independently by Wiz. The company also released a PHP vulnerability checker and proposed a framework for coordinating responses to AI-discovered security flaws.” via TLDR AI

    Reported byThe Rundown AI, Superhuman AI, TLDR AI, AI Breakfast

    Discussed onHacker News 266 pts · 183 comments

    How 1 source put it
    • The Rundown AI2026-10-22issue ↗
    • Superhuman AI2026-10-22issue ↗
    • TLDR AI2026-10-22issue ↗ OpenAI coordinated with PHP maintainers and security partners to patch critical vulnerabilities discovered using GPT-5.6 Sol and independently by Wiz. The company also released a PHP vulnerability checker and proposed a framework for coordinating responses to AI-discovered security flaws.
    • AI Breakfast2026-10-23issue ↗
    openai.com First shared 2026-10-21
  6. Shared by 3

    Introducing Flint

    Recursive launches Flint, a 40 billion parameter MoE coding model built in under four months. Includes Apache 2.0 weights, rec inference engine, and benchmarks showing 125 tokens per second on an M4 Pro MacBook Pro. cited

    Reported byThe Rundown AI, TLDR AI, Latent Space / AINews

    Where it was shared
    recursive.com announcement · 5 min read First shared 2026-10-22
  7. Shared by 7

    Investing in the public benefit

    Anthropic commits 5% of shares to a new Public Benefit Fund, adding 2% for every 50% valuation increase. Details how the LLC avoids lobbying, funds labor impact studies, and combines with founder pledges to reach 20% total public good equity. cited

    Reported byThe Rundown AI, The Neuron, Superhuman AI, TLDR AI, AI Breakfast, Latent Space / AINews, Forward Future

    Where it was shared
    anthropic.com announcement · 2 min read First shared 2026-10-19
  8. Shared by 4

    deepseek-ai/DeepInfer: DeepInfer: High-performance, large-scale inference framework for DeepSeek models

    DeepSeek releases DeepInfer, an open-source inference framework supporting DeepSeek-V4.1-Flash on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs. Includes configuration for DSpark verification, SWA bounded replay, and batch-invariant inference, plus build instructions requiring Python 3.12 and CUDA 13. cited

    Reported byInterconnects, Latent Space / AINews, The Rundown AI

    Discussed onHacker News 2,346 pts · 891 comments

    Where it was shared
    github.com repo · 2 min read First shared 2026-10-20
  9. Shared by 2

    OpenAI nears up to $30 billion funding round at $1.25 trillion valuation led by UAE investors

    OpenAI raises up to $30 billion at a $1.25 trillion valuation led by United Arab Emirates investors. Includes a commercial deal to secure more computing capacity from the UAE alongside the new investment. cited

    Reported byThe Rundown AI, TLDR AI

    Where it was shared
    cnbc.com news report · 1 min read First shared 2026-10-23
  10. Shared by 4

    OpenAI Acquires Orchestra Bio to Accelerate Drug Development

    “OpenAI acquired Orchestra Bio, whose agents flag how a delay in one part of a drug program ripples through the rest, and will fold it into its Rosalind life-science tools.” via The Neuron

    Reported byThe Neuron, The Rundown AI, TLDR AI, AI Breakfast

    How 1 source put it
    openai.com First shared 2026-10-21