1
Amazon saves millions by replacing Sonnet 4.6 with Strands Decider
Two Amazon engineers swapped Claude Sonnet 4.6 for a decision model in one of the world's biggest internal ticketing systems.
New
The latest blog posts to take off on Hacker News. 3,236 posts.
2
TPU 8tx and TPU 8i Max
We’ve expanded our eighth generation TPU portfolio with TPU 8tx and TPU 8i Max, marking four new TPU systems in a single year.
3
Claude Fable 5.2 scores 65 on Artificial Analysis, highest so far
Analysis of Anthropic's Claude Fable 5.2 (Max, Default Fallback) and comparison to other AI models across key metrics including quality, price, performance (time per Intelligence Index task & time to first answer token), context window & more.
4
Claude Fable 5.2 and Mythos 5.2
Claude Fable 5.2 and Claude Mythos 5.2, our new models for the hardest technical challenges. Fable 5.2 sets a new bar in coding, finance, and the sciences.
5
Muse Ember: Our smallest, most efficient model
Muse Ember is our smallest and most efficient model, designed for complex reasoning execution and the work agents repeat all day.
6
Apple capped engineers at a $200 daily AI budget
The world's richest company is suddenly rationing compute for its own builders.
7
Statement from Nvidia board of directors
The NVIDIA Board is initiating an independent investigation in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort. A committee of ...
8
Agent memory needs its OpenTelemetry moment
Every agent tool stores memory in its own shape, and every week someone builds another one. None of them can read the others. Observability had the same problem until OpenTelemetry gave everyone one data model to share. A tour of today's memory formats, the proposals that already exist, and what a shared spec could…
9
Claude Code's "Don't Ask Again" exposes passwords in 1,122 public GitHub files
Secure everything coding agents build, use, and ship - prevent risk before code exists, AutoFix the backlog, and protect the coding agents and supply chain.
10
MaxxBench – "Nice DeepSWE score. But is it true?"
How coding models respond when familiar engineering tasks ask for something different.
11
Every Mac comes with an (undocumented) family member classifier
Every Mac since macOS 13 (2022) ships a pair of daemons whose only purpose in life is to build a graph about who you are, who you know, how those people are related to you and to each other, how you communicate with them, where you go, what you do and when. In this post, we look into how this system works and give you…
12
The Components of Agent Memory
For Claude to learn from experience, it has to decide what’s worth remembering before it knows what’s next. Here’s how “dreaming” helps it organize memory and correct mistakes.
13
The AI sell-off compared to well-known stock market crashes
The S&P 500 fell 4.4% on 5 Nov 2026, its worst day since April 2025, and is 6% below its 6 Oct record. Charted day by day against 1987, 2008, Covid and other falls since 1950.
14
A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost
We tested Jev and GPT-6 Luna as judges for identity resolution. A hybrid of Jev and Claude Opus matched frontier accuracy at over 4x lower cost.
15
Gemini Omni 1.2 Flash
Gemini Omni 1.2 Flash brings native 15-second video generation and editing, stronger consistency, and 30% lower pricing.
16
AI for scientific discovery: How researchers use Claude
We analyzed 207,000 research sessions to learn how Claude is used in science, why some tasks get more use than others, and how AI could speed up discovery.
17
Inkling 2 Small
Today we are introducing the second generation of Inkling models with the release of Inkling 2 Small.
18
Baidu released ERNIE 5.5 Flash Base
ERNIE 5.5 Flash Base: a 280B-A14B hybrid-attention MoE pretrained on 33.5T tokens. An open, efficient pretraining base.
19
GPT-6 Astra covertly underperforms on safety research
Research on aligning AI with human values and intent, and reports documenting model failures.
20
Images 2.6
Images 2.6 brings faster image generation, GIF creation, more precise editing, and new API controls across ChatGPT, Codex, and the API.
21
A thousand fewer paths to root
OpenAI built an autonomous pipeline that found, validated, and generated patches for 948 local privilege-escalation vulnerabilities in the Linux kernel.
22
Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
We develop frontier AI methods to reinvent chip development, collapsing traditional design timelines, closing the loop between AI and the hardware that fuels it, and enabling a Cambrian explosion of custom silicon.
23
Progress on provable inference for large language models
Sharing progress on a powerful type of cryptographic proof known as a zkSNARK, which is uniquely fit for provable inference.
24
Nvidia RTX Spark encrypted firmware slows inference speed to 35% on Linux
A hands-on look at dual-booting Linux on the Nvidia RTX Spark laptop, the GPU clock throttling bug found, and why WSL still wins.
25
There is no posttraining data commons
The LLM revolution ran on a pretraining commons that humanity built for other reasons. Posttraining consumes tasks, not documents, and no such commons exists. If the open models community wants...
26
China's First Frontier AI Export Controls: Mofcom Announcement No. 48
China’s First Frontier AI Export Controls: MOFCOM Announcement No. 48 Key Takeaways On October 16, 2026, China’s Ministry of Commerce (MOFCOM) and Ministry of Science and Technology (MO…
27
We're training a Figma-fluent editing model
Here, we’re sharing the research behind how we’re translating that to design editing workflows.
28
Malware uses LLMs to rank victims by how much they're worth
The self-deleting VOLTA infostealer steals credentials, sessions, and cryptocurrency wallets. Then, it uses LLMs to profile victims and rank their most valuable assets.
29
Do the math: How we took over ChatGPT accounts
By bypassing ChatGPT's sandbox and content security policy, our research team was able to take over anyone's account simply by sending them a link.
30
AI isn't replacing SREs, it's deskilling them
When AI handles 95% of your incident response, do you get worse at handling the 5% that actually matters?