Modern Web Guidance 1.0: The web platform expertise your coding agent is missing
7 points and 0 comments on Hacker News, linking developer.chrome.com.
Top now
6 storiesWhat is climbing fastest across every source
NVIDIA6.9xJensen Huang6.9xUnited Airlines8.2xFacebook9.9xMACaQ9.3xGemini 4 Carbon6.3xCEO Jensen Huang8.6xaion-2-trainer-tool7.4xCEO6.6xClaude5.5x
Jensen Huang Said to Have Had Relationship with Nvidia Executive
80 points and 48 comments on Hacker News, linking nytimes.com.
facebook/sam-3d-body-dinov3
A model model from facebook with 1,001 likes and 14.4k downloads on Hugging Face.
Gilbert + Tobin rethinks the business of law with AI
Gilbert + Tobin rethinks the business of law with AI in a conversation with Sam Nickless, Partner and Chief Executive Officer. ·
openai/codex 0.167.2
OpenAI releases Codex version 0.167.2, which sends tool updates incrementally by default on Google Cloud Vertex AI. ·
Gemini Decisions API
The Gemini Decisions API uses the Gemini Decision model to provide fast, calibrated multimodal decisions. ·
216 points and 73 comments on Hacker News, linking blog.google.
Jensen Huang Said to Have Had Relationship with Nvidia Executive
80 points and 48 comments on Hacker News, linking nytimes.com.
facebook/sam-3d-body-dinov3
A model model from facebook with 1,001 likes and 14.4k downloads on Hugging Face.
facebook/dinov3-vitl16-pretrain-lvd1689m
Facebook releases DINOv3, an image feature extraction model with 0.3 billion parameters available on Hugging Face. ·
unslothai/unsloth Dynamic v3.1 GGUFs + Performance Fixes
Unsloth releases Qwen3.8-27B Dynamic v3.1 GGUFs with performance fixes, smoother rendering, and new coding agent features. ·
langchain==1.5.0
Langchain 1.5.0 drops Python 3.10 support, requiring Python 3.11 and langchain-core 1.8.0. ·
Gemini Decisions API
The Gemini Decisions API uses the Gemini Decision model to provide fast, calibrated multimodal decisions. ·
216 points and 73 comments on Hacker News, linking blog.google.
Gemini 4 Carbon is now available
291 points and 164 comments on Hacker News, linking blog.google.
Dodaios/ste80-humanizer
A new repository written in Python with 404 stars in its first week.
unslothai/unsloth Dynamic v3.1 GGUFs + Performance Fixes
Unsloth releases Qwen3.8-27B Dynamic v3.1 GGUFs with performance fixes, smoother rendering, and new coding agent features. ·
dimensiondriverarbor/aion-2-trainer-tool
Aion 2 is an AI-driven software suite for advanced prompt engineering tasks that offers a pre-configured environment. ·
Tree-Ising: Repairing Language Models with Bit-Level Ising Edits
Tree-Ising repairs Qwen3.5-0.8B reasoning loops by solving sparse discrete optimization problems over BF16 weights, achieving 54.6% on MATH-500. ·
Trending topics today: heat
Table
| NVIDIA | 3.8 |
| Jensen Huang | 2.13 |
| United Airlines | 1.75 |
| 1.59 | |
| MACaQ | 1.12 |
| Gemini 4 Carbon | 1.09 |
Heat = summed hotness of today's stories naming the topic.
Models
5 storiesReleases, papers, trending weights and new listings
Top 6 trending on Hugging Face: likes
Table
| Lightricks/LTX-2.5 ▼1 | 11510 |
| xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic | 3002 |
| speridlabs/idol ▼1 | 2143 |
| moonshotai/Kimi-K3.1-Pro ▼2 | 1934 |
| MiniMaxAI/MiniMax-H3-Image ▲5 | 697 |
| NousResearch/Arete-27B ▲2 | 434 |
PirateBench: score out of 100
Table
| 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/
Gemini Decisions API
The Gemini Decisions API uses the Gemini Decision model to provide fast, calibrated multimodal decisions. ·
216 points and 73 comments on Hacker News, linking blog.google.
Huawei releases MACaQ, its internal quantizer for local inference
Huawei releases MACaQ, an internal quantizer for local inference that supports model-agnostic quantization for GGUF, local inference and beyond. ·
Tools for exploring the souls of large language models
TinyGPT Visualizer explores the metaphysics and souls of large language models to decode how they think and act. ·
facebook/dinov3-vitl16-pretrain-lvd1689m
Facebook releases DINOv3, an image feature extraction model with 0.3 billion parameters available on Hugging Face. ·
Launch HN: Pomerado (YC F26) – Turn any website into an MCP for agents
36 points and 32 comments on Hacker News, linking pomerado.ai.
Gemini Decisions API
The Gemini Decisions API uses the Gemini Decision model to provide fast, calibrated multimodal decisions. ·
216 points and 73 comments on Hacker News, linking blog.google.
Gemini 4 Carbon is now available
291 points and 164 comments on Hacker News, linking blog.google.
Tree-Ising: Repairing Language Models with Bit-Level Ising Edits
Tree-Ising repairs Qwen3.5-0.8B reasoning loops by solving sparse discrete optimization problems over BF16 weights, achieving 54.6% on MATH-500. ·
Muse Ember: Our smallest, most efficient model
332 points and 140 comments on Hacker News, linking research.meta.ai.
World In-Context Learning for Agentic Robot Manipulation
RoboScript implements World In-Context Learning for agentic robot manipulation, achieving 83.4% mean success on 50 RoboTwin 2.0 tasks. ·
X voices
5 storiesX posts other sources are linking to, then named voices on Bluesky and Mastodon
@watcherguru
Nvidia prepares to fire CEO Jensen Huang, WSJ reports
111 points and 36 comments on Hacker News, linking an X post.
@dimitrispapail
Reprogramming Dormant Neurons to Backdoor AI Agents
6 points and 1 comments on Hacker News, linking an X post.
@zulali
Muhammad on X: "Claude reverse engineered my 32-year-old Jeep
2 points and 0 comments on Hacker News, linking an X post.
@TheMidasProj
Fake accounts defended Jensen Huang and attacked lawmakers over chip exports
8 points and 0 comments on Hacker News, linking an X post.
@ericweinstein
AI assisted theoretical physics paper intercepted by "shepherding committee."
1 points and 0 comments on Hacker News, linking an X post.
@watcherguru
Nvidia prepares to fire CEO Jensen Huang, WSJ reports
111 points and 36 comments on Hacker News, linking an X post.
@dimitrispapail
Reprogramming Dormant Neurons to Backdoor AI Agents
6 points and 1 comments on Hacker News, linking an X post.
@TheMidasProj
Fake accounts defended Jensen Huang and attacked lawmakers over chip exports
8 points and 0 comments on Hacker News, linking an X post.
@willdepue
We just found a new Dwarf Planet in our solar system with Claude and you can too
3 points and 0 comments on Hacker News, linking an X post.
@sundarpichai
Gemini 4 Carbon is rolling out today
14 points and 0 comments on Hacker News, linking an X post.
GitHub
5 storiesNew repos, releases and changelog
New AI repos this week: stars
Table
| DIYgod/pinky | 1859 |
| voyager-so/vork | 1435 |
| Arda313131882/aion-2-trainer-tool | 1296 |
| HybridRobotics/supermimic | 1019 |
| JosssphZhou/poki | 972 |
| majorminorlabs/hermes-voice | 673 |
Huawei releases MACaQ, its internal quantizer for local inference
Huawei releases MACaQ, an internal quantizer for local inference that supports model-agnostic quantization for GGUF, local inference and beyond. ·
46 points and 18 comments on Hacker News, linking github.com.
unslothai/unsloth Dynamic v3.1 GGUFs + Performance Fixes
Unsloth releases Qwen3.8-27B Dynamic v3.1 GGUFs with performance fixes, smoother rendering, and new coding agent features. ·
Show HN: Klad, a paint app your coding agent can drive (source-available)
60 points and 26 comments on Hacker News, linking github.com.
langchain==1.5.0
Langchain 1.5.0 drops Python 3.10 support, requiring Python 3.11 and langchain-core 1.8.0. ·
dimensiondriverarbor/aion-2-trainer-tool
Aion 2 is an AI-driven software suite for advanced prompt engineering tasks that offers a pre-configured environment. ·
Dodaios/ste80-humanizer
A new repository written in Python with 404 stars in its first week.
unslothai/unsloth Dynamic v3.1 GGUFs + Performance Fixes
Unsloth releases Qwen3.8-27B Dynamic v3.1 GGUFs with performance fixes, smoother rendering, and new coding agent features. ·
dimensiondriverarbor/aion-2-trainer-tool
Aion 2 is an AI-driven software suite for advanced prompt engineering tasks that offers a pre-configured environment. ·
Show HN: Klad, a paint app your coding agent can drive (source-available)
60 points and 26 comments on Hacker News, linking github.com.
Huawei releases MACaQ, its internal quantizer for local inference
Huawei releases MACaQ, an internal quantizer for local inference that supports model-agnostic quantization for GGUF, local inference and beyond. ·
Infrastructure
5 storiesData centers, power, chips and what tokens cost
Data-center and energy stories per day
Table
| Fri 06 | 25 |
| Sat 07 | 1 |
| Sun 08 | 2 |
| Mon 09 | 35 |
| Tue 10 | 31 |
| Wed 11 | 31 |
| Thu 12 | 0 |
Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.
Output price, $ per million tokens (newest models)
Table
| ernie-light-dt-1.0-intl | 3.94 |
| gemini-4-carbon | 10 |
| gemini-4-carbon:batch | 5 |
| nex-n2.6 | 0.4 |
| claude-fable-5.2 | 40 |
| median, all paid models | 2 |
441 paid models listed; median of the 37 added in 30 days: $2.50. Rankings data by OpenRouter, CC BY 4.0.
A false smoke alarm almost drowned Jane Street's only NYC datacenter
383 points and 189 comments on Hacker News, linking michael.bacarella.com.
Consortium to develop 400MW hyperscale AI data center in Oman
A consortium plans to build a 400MW hyperscale AI data center in Oman, with US firm Crusoe interested in purchasing capacity. ·
Working hard Evri day
The article describes the digital infrastructure supporting a major UK courier service and details the background of its chief technology officer. ·
A false smoke alarm almost drowned Jane Street's only NYC datacenter
383 points and 189 comments on Hacker News, linking michael.bacarella.com.
TPU 8tx and TPU 8i Max
258 points and 76 comments on Hacker News, linking blog.google.
The same H100 costs $2.69 or $12.29 an hour depending where you rent it
171 points and 86 comments on Hacker News, linking github.com.
Compress any LLM 3x in one click,near-lossless,in under 50 MB RAM
SPECTRA compresses large language models three times with near-lossless accuracy using under 50 megabytes of RAM and achieves 82.53 tokens per second on 4GB consumer GPUs. ·
Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU
Llama-modchip runs DeepSeek-V4-Flash and 30B MoEs on 64GB RAM without GPUs, achieving up to 111 tokens per second. ·