MiniMaxAI/MiniMax-H3
MiniMax-H3 is a 33B parameter image-text-to-video model available on Hugging Face. ·
A image text to video model from MiniMaxAI with 6,587 likes and 6.4M downloads on Hugging Face.
Top now
6 storiesWhat is climbing fastest across every source
OpenAI6.5xAnthropic5.4xGoogle6.8xLLMs6.8xRipple-65B Preview6.9xAMD7.3xNvidia6.7xMultimodal Understanding6.0xQwen 3.8 27B5.9xAWS5.5x
You'll need nerves of steel to work with AI agents
23 points and 20 comments on Hacker News, linking orchidfiles.com.
AI poisoning: are we using AI, or is AI using us?
AI poisoning explores how artificial intelligence provides an illusion of unlimited power while quietly changing human thinking, creation, and time usage. ·
openai/openai-python v3.46.0
OpenAI releases version 3.46.0 of its Python library, adding GPT-6 Rosalind and residency options for Australia, India, and Japan. ·
langchain-core==1.7.6
New release langchain-core==1.7.6 of langchain-ai/langchain on GitHub.
PrunaAI/P-H3-Image-Comfy
PrunaAI releases P-H3-Image-Comfy, a text-to-image diffusion model available as a single file on Hugging Face. ·
moonshotai/Kimi-K3.1-Pro
Kimi K3.1 Pro is an image-text-to-text model with 2.8 trillion parameters available on Hugging Face. ·
376 points and 102 comments on Hacker News, linking huggingface.co.
MiniMaxAI/MiniMax-H3
MiniMax-H3 is a 33B parameter image-text-to-video model available on Hugging Face. ·
OpenAI to acquire Pierre Computer Company
OpenAI acquires Pierre Computer Company to expand Codex tools for building, understanding, and reviewing code. ·
You'll need nerves of steel to work with AI agents
23 points and 20 comments on Hacker News, linking orchidfiles.com.
unslothai/unsloth Improved Customization + Glass + Browser Use
Unsloth releases Qwen3.8-Flash-Next, updates Glass themes, adds a Composer position slider, and enables browser file interaction with lower idle memory usage. ·
Balanov/latch
Balanov/latch is a tiny HTTPS contract for agents and services featuring quote, approve, commit, and receipt steps with zero dependencies under the MIT license. ·
Kimi K3.1 Pro model weights are out
Kimi K3.1 Pro is an image-text-to-text model with 2.8 trillion parameters available on Hugging Face. ·
376 points and 102 comments on Hacker News, linking huggingface.co.
OpenAI to acquire Pierre Computer Company
OpenAI acquires Pierre Computer Company to expand Codex tools for building, understanding, and reviewing code. ·
MiniMaxAI/MiniMax-H3
MiniMax-H3 is a 33B parameter image-text-to-video model available on Hugging Face. ·
ggml-org/llama.cpp v0.9.0
ggml-org releases llama.cpp v0.9.0, adding OCP FP8 E4M3 quantization, flash attention support, and four new model architectures. ·
unslothai/unsloth Improved Customization + Glass + Browser Use
Unsloth releases Qwen3.8-Flash-Next, updates Glass themes, adds a Composer position slider, and enables browser file interaction with lower idle memory usage. ·
worldonlyu/Google_AI_Pro_2Year_ZeroCost_Auto
A new repository written in Python with 548 stars in its first week.
Trending topics today: heat
Table
| OpenAI | 2.22 |
| Anthropic | 1.87 |
| 1.79 | |
| LLMs | 0.99 |
| Ripple-65B Preview | 0.92 |
| AMD | 0.81 |
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
| deepseek-ai/DeepSeek-V4.1-Flash-1017 ▲1 | 5271 |
| mistralai/Mistral-Large-4-1T-A52B ▼1 | 1591 |
| nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B ▲1 | 1492 |
| LiquidAI/LFM3-1.7B ▼3 | 1485 |
| speridlabs/idol ▲2 | 619 |
| moonshotai/Kimi-K3.1-Pro new | 514 |
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/
moonshotai/Kimi-K3.1-Pro
Kimi K3.1 Pro is an image-text-to-text model with 2.8 trillion parameters available on Hugging Face. ·
376 points and 102 comments on Hacker News, linking huggingface.co.
Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
Ripple-65B is a U.S. open-weight model that achieves state-of-the-art performance on agentic coding tasks. ·
Progress on provable inference for large language models
The item shares progress on zkSNARKs, a type of cryptographic proof suitable for provable inference in large language models. ·
Launch HN: Kino AI (YC S23) – AI media manager and video editor
70 points and 26 comments on Hacker News, linking news.ycombinator.com.
Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)
628 points and 163 comments on Hacker News, linking news.ycombinator.com.
Kimi K3.1 Pro model weights are out
Kimi K3.1 Pro is an image-text-to-text model with 2.8 trillion parameters available on Hugging Face. ·
376 points and 102 comments on Hacker News, linking huggingface.co.
Qwen 3.8 27B on my laptop can interpret model weights, too
A paper by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab describes weight-space programming, where models like GPT-6 Astra read and write neural network weights. ·
Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)
628 points and 163 comments on Hacker News, linking news.ycombinator.com.
Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
Ripple-65B is a U.S. open-weight model that achieves state-of-the-art performance on agentic coding tasks. ·
Progress on provable inference for large language models
The item shares progress on zkSNARKs, a type of cryptographic proof suitable for provable inference in large language models. ·
X voices
5 storiesX posts other sources are linking to, then named voices on Bluesky and Mastodon
@josharosen
Agent State Machines: Building Deterministic Scaffolding for Agents
1 points and 0 comments on Hacker News, linking an X post.
@RepYassAnsari
Yassamin Ansari: Top of the Democratic agenda on day 1 pause on advanced AI dev
2 points and 6 comments on Hacker News, linking an X post.
@ClaudeDevs
Background subagents in Claude Code now use less of your usage limit
2 points and 0 comments on Hacker News, linking an X post.
@frostedxyz
Researcher accuses Anthropic of stealing his work
8 points and 1 comments on Hacker News, linking an X post.
@pierrecomputer
Pierre Computer is being acquired by OpenAI
72 points and 32 comments on Hacker News, linking an X post.
@pierrecomputer
Pierre Computer is being acquired by OpenAI
72 points and 32 comments on Hacker News, linking an X post.
@RepYassAnsari
Yassamin Ansari: Top of the Democratic agenda on day 1 pause on advanced AI dev
2 points and 6 comments on Hacker News, linking an X post.
@frostedxyz
Researcher accuses Anthropic of stealing his work
8 points and 1 comments on Hacker News, linking an X post.
@ns123abc
Railway employee's agent caught hacking competitors
33 points and 6 comments on Hacker News, linking an X post.
@ONagel33303
Abdul El-Sayed is the Pause AI candidate: Chinese labs are just spoofing off US
5 points and 1 comments on Hacker News, linking an X post.
GitHub
5 storiesNew repos, releases and changelog
New AI repos this week: stars
Table
| deepseek-ai/dsh-superagent-prototype | 18587 |
| KKKKhazix/agent-autoloop | 2526 |
| caru-ini/stack-diagram | 1752 |
| wquguru/cicada | 1144 |
| N1nEmAn/sapper-v2 | 858 |
| anthropics/FlashProve | 790 |
unslothai/unsloth Improved Customization + Glass + Browser Use
Unsloth releases Qwen3.8-Flash-Next, updates Glass themes, adds a Composer position slider, and enables browser file interaction with lower idle memory usage. ·
New release v0.1.915-beta of unslothai/unsloth on GitHub.
Balanov/latch
Balanov/latch is a tiny HTTPS contract for agents and services featuring quote, approve, commit, and receipt steps with zero dependencies under the MIT license. ·
worldonlyu/Google_AI_Pro_2Year_ZeroCost_Auto
A new repository written in Python with 548 stars in its first week.
ggml-org/llama.cpp v0.9.0
ggml-org releases llama.cpp v0.9.0, adding OCP FP8 E4M3 quantization, flash attention support, and four new model architectures. ·
anthropics/claude-code v2.1.318
Anthropic releases Claude Code version 2.1.318, which adds mod answer records, assume role support, and fixes for macOS session stalls and permission prompts. ·
ggml-org/llama.cpp v0.9.0
ggml-org releases llama.cpp v0.9.0, adding OCP FP8 E4M3 quantization, flash attention support, and four new model architectures. ·
New release v0.9.0 of ggml-org/llama.cpp on GitHub.
unslothai/unsloth Improved Customization + Glass + Browser Use
Unsloth releases Qwen3.8-Flash-Next, updates Glass themes, adds a Composer position slider, and enables browser file interaction with lower idle memory usage. ·
worldonlyu/Google_AI_Pro_2Year_ZeroCost_Auto
A new repository written in Python with 548 stars in its first week.
anthropics/claude-code v2.1.318
Anthropic releases Claude Code version 2.1.318, which adds mod answer records, assume role support, and fixes for macOS session stalls and permission prompts. ·
Postman silently auto-installs AI plugins and attempts to share all API data
Postman silently auto-installs AI plugins and attempts to share all API data with users. ·
Infrastructure
5 storiesData centers, power, chips and what tokens cost
Data-center and energy stories per day
Table
| Fri 30 | 29 |
| Sat 31 | 2 |
| Sun 01 | 5 |
| Mon 02 | 34 |
| Tue 03 | 39 |
| Wed 04 | 40 |
| Thu 05 | 0 |
Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.
Median output price of models added in the last 30 days, $ per million tokens
One point per OpenRouter reading, last 8 days. Data by OpenRouter, CC BY 4.0.
Output price, $ per million tokens (newest models)
Table
| ripple-65b-preview | 0.6 |
| hy4 | 1.2505 |
| glm-5.3-flashx-legal | 1.25 |
| particle-1 | 0.5 |
| gpt-6.1-terra-pro | 5 |
| median, all paid models | 2 |
429 paid models listed; median of the 31 added in 30 days: $2.50. Rankings data by OpenRouter, CC BY 4.0.
Qualcomm Cloud AI 100 Ultra – 150W AI Accelerator Card with 128GB at 548GB/s
Qualcomm Cloud AI 100 Ultra is a 150W AI accelerator card with 128GB memory and 548GB/s bandwidth that accelerates generative AI. ·
100 points and 43 comments on Hacker News, linking qualcomm.com.
US voters roundly reject data centers in ballot measures
US voters in over 30 localities across nine states reject data centers through ballot measures including moratoriums and bans. ·
RTX 5090 Street Prices Top $9k as Nvidia GPU Costs Climb
Nvidia RTX 5090 street prices exceed $9,000 in the U.S., representing a 370% increase from launch, while other RTX 50 series cards rise due to a memory shortage. ·
AI stocks slide as Texas governor result raises data center risks
Tech stocks decline in overnight trading as the probability of a blue wave increases following the Texas governor result. ·
US voters roundly reject data centers in ballot measures
US voters in over 30 localities across nine states reject data centers through ballot measures including moratoriums and bans. ·
308 points and 620 comments on Hacker News, linking semafor.com.
Qualcomm Cloud AI 100 Ultra – 150W AI Accelerator Card with 128GB at 548GB/s
Qualcomm Cloud AI 100 Ultra is a 150W AI accelerator card with 128GB memory and 548GB/s bandwidth that accelerates generative AI. ·
AI stocks slide as Texas governor result raises data center risks
Tech stocks decline in overnight trading as the probability of a blue wave increases following the Texas governor result. ·
RTX 5090 Street Prices Top $9k as Nvidia GPU Costs Climb
Nvidia RTX 5090 street prices exceed $9,000 in the U.S., representing a 370% increase from launch, while other RTX 50 series cards rise due to a memory shortage. ·