The front page of the internet for AI users · Thu 15 Oct 2026, 02:00 UTC · Get it by email
What is climbing fastest across every source
LFM2.5-24B-A2B is an open-weight reasoning model with 24 billion total and 2 billion active parameters designed for fast, private on-device agents. It outperforms Qwen3.6-35B-A3B on LiveClaw. ·
73 points and 18 comments on Hacker News, linking liquid.ai.
The text introduces a new open-source model that accepts remixing but rejects pull requests, referencing historical software development concepts. ·
From Gary Marcus.
OpenAI and Anthropic discussed how to grant financial agents useful autonomy while keeping consequential decisions under control at the AI Engineer New York event. ·
From Simon Willison.
New release v2.1.300 of anthropics/claude-code on GitHub.
LFM2.5-24B-A2B is an open-weight reasoning model with 24 billion total and 2 billion active parameters designed for fast, private on-device agents. It outperforms Qwen3.6-35B-A3B on LiveClaw. ·
73 points and 18 comments on Hacker News, linking liquid.ai.
258 points and 112 comments on Hacker News, linking sunshowers.io.
New release v2.1.300 of anthropics/claude-code on GitHub.
A new repository written in Swift with 198 stars in its first week.
Graphicle is a Rust program that provides code graph and memory features for coding agents, supporting Claude Code, Codex, Cursor, and any MCP client. ·
A text to speech model from fishaudio with 317 likes and 839 downloads on Hugging Face.
New release v2.1.300 of anthropics/claude-code on GitHub.
Anthropic built Scion, a custom source control system, after daily code merges rose from 2,500 in May 2026 to over 30,000 by September. ·
Security specialists attribute recent AI hacking incidents to containment and monitoring failures in labs, rather than autonomous agents, and advise CISOs to vet vendors and strengthen security basics. ·
New release v2.1.299 of anthropics/claude-code on GitHub.
VideoJev adapts multimodal language models to treat video understanding as structured decision tasks, improving mean mIoU by up to 10.8 points over its backbones. ·
Jeva is a decision-native architecture that addresses structural requirements for low-latency models by using a shared encoder and compact decision stack to handle variable candidate sets and null outcomes. ·
| Anthropic | 2.74 |
| LLM | 2.16 |
| OpenAI | 2.09 |
| LFM2.5 | 1.27 |
| LFM2.5-24B-A2B | 1.27 |
| Built | 1.05 |
Heat = summed hotness of today's stories naming the topic.
Releases, papers, trending weights and new listings
| abenzerps/Qwen-Image-2.1-Uncensored-Quantized ▼1 | 4309 |
| google/embeddinggemma-2 | 2266 |
| jialinyyzz/humanizer | 1638 |
| Qwen/Qwen-Image-2.1-Turbo | 776 |
| autotrust/JEV-35B ▲6 | 731 |
| openbmb/MiniCPM-V-4.7-35B-A3B ▲2 | 626 |
| gpt-6.1-sol | 98.9 |
| opus-5-5 | 96.46 |
| gpt-6-sol | 91.6 |
| gpt-6-luna | 65.25 |
| haiku-4-5 | 4.28 |
Agents build our own projects from a brief; every run, failures included: https://cowerx.dev/daily/piratebench/
A text to speech model from fishaudio with 317 likes and 839 downloads on Hugging Face.
VideoJev adapts multimodal language models to treat video understanding as structured decision tasks, improving mean mIoU by up to 10.8 points over its backbones. ·
Jeva is a decision-native architecture that addresses structural requirements for low-latency models by using a shared encoder and compact decision stack to handle variable candidate sets and null outcomes. ·
Vela is a framework that converts real user behavior data into training tasks for large language models. It uses records from 100,000 users to generate 100,000 tasks across 91 templates. ·
Agent-Reflex is a training framework that helps language agents reflect on noisy tool feedback and user corrections to improve their performance during long-horizon reasoning tasks. ·
230 points and 55 comments on Hacker News, linking developers.googleblog.com.
VideoJev adapts multimodal language models to treat video understanding as structured decision tasks, improving mean mIoU by up to 10.8 points over its backbones. ·
Jeva is a decision-native architecture that addresses structural requirements for low-latency models by using a shared encoder and compact decision stack to handle variable candidate sets and null outcomes. ·
Vela is a framework that converts real user behavior data into training tasks for large language models. It uses records from 100,000 users to generate 100,000 tasks across 91 templates. ·
Agent-Reflex is a training framework that helps language agents reflect on noisy tool feedback and user corrections to improve their performance during long-horizon reasoning tasks. ·
X posts other sources are linking to, then named voices on Bluesky and Mastodon
Can Coding Agents Act Like Staff Engineers?
2 points and 1 comments on Hacker News, linking an X post.
I spent 48 hours in SF hanging out with billionaires, data labs and AI execs
3 points and 0 comments on Hacker News, linking an X post.
AI researchers are taking up smoking to signal to other researchers
3 points and 0 comments on Hacker News, linking an X post.
FSD Supervised v14 Lite now rolling out to Intel Model S and X
2 points and 0 comments on Hacker News, linking an X post.
I just got our entire company's Claude Enterprise account suspended
14 points and 3 comments on Hacker News, linking an X post.
Can Coding Agents Act Like Staff Engineers?
2 points and 1 comments on Hacker News, linking an X post.
I spent 48 hours in SF hanging out with billionaires, data labs and AI execs
3 points and 0 comments on Hacker News, linking an X post.
AI researchers are taking up smoking to signal to other researchers
3 points and 0 comments on Hacker News, linking an X post.
OpenAI finds recording device in one of their offices
85 points and 2 comments on Hacker News, linking an X post.
I just got our entire company's Claude Enterprise account suspended
14 points and 3 comments on Hacker News, linking an X post.
New repos, releases and changelog
| onehumanbeing/esp32-iphone-agent | 2097 |
| rociiu/talorys | 1002 |
| zhuyansen/new-term-radar | 963 |
| marclou/mailcheap | 933 |
| jwuthri-cloud/Tracely-ai | 817 |
| outcrawlai/scout | 657 |
Neckbeard is a development operations plugin that scaffolds tested cloud infrastructure from a versioned module catalog. ·
56 points and 47 comments on Hacker News, linking github.com.
Graphicle is a Rust program that provides code graph and memory features for coding agents, supporting Claude Code, Codex, Cursor, and any MCP client. ·
A new repository written in Swift with 198 stars in its first week.
Graphicle is a Rust program that provides code graph and memory features for coding agents, supporting Claude Code, Codex, Cursor, and any MCP client. ·
New release v2.1.299 of anthropics/claude-code on GitHub.
Neckbeard is a development operations plugin that scaffolds tested cloud infrastructure from a versioned module catalog. ·
56 points and 47 comments on Hacker News, linking github.com.
New release v2.1.299 of anthropics/claude-code on GitHub.
A new repository written in Swift with 198 stars in its first week.
Graphicle is a Rust program that provides code graph and memory features for coding agents, supporting Claude Code, Codex, Cursor, and any MCP client. ·
Graphicle is a Rust program that provides code graph and memory features for coding agents, supporting Claude Code, Codex, Cursor, and any MCP client. ·
Data centers, power, chips and what tokens cost
| Fri 09 | 30 |
| Sat 10 | 3 |
| Sun 11 | 7 |
| Mon 12 | 41 |
| Tue 13 | 29 |
| Wed 14 | 34 |
| Thu 15 | 0 |
Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.
One point per OpenRouter reading, last 8 days. Data by OpenRouter, CC BY 4.0.
| pareto-latest | 7.5 |
| pareto-26.9 | 7.5 |
| lfm-2.5-24b-a2b | 0.7 |
| gemini-4-argon | 10 |
| gemini-4-argon:batch | 5 |
| median, all paid models | 2 |
436 paid models listed; median of the 44 added in 30 days: $5.00. Rankings data by OpenRouter, CC BY 4.0.
290 points and 216 comments on Hacker News, linking startupfortune.com.
60 points and 39 comments on Hacker News, linking dfarq.homeip.net.
Brazil expands data center construction to Rio, Fortaleza, and the south, while grid delays and permitting processes determine which regional markets succeed. ·
Panduit's Hans Obermillacher discusses AI demand surges, retail edge computing growth, Nordic geo-shifting, and evolving data center infrastructure needs. ·
290 points and 216 comments on Hacker News, linking startupfortune.com.
60 points and 39 comments on Hacker News, linking dfarq.homeip.net.
25 points and 39 comments on Hacker News, linking reuters.com.
Small AI models may soon self-replicate across cheap IoT devices like Android TV boxes without requiring a datacenter. ·
Researchers describe a distributed virtual GPU cloud that utilizes idle Apple Silicon devices to address limitations in centralized machine learning inference infrastructure. ·
