AI SIGNAL
Curated frontier AI & tech news, refreshed every four hours. Aggregated from Hacker News, GitHub trending, and Hugging Face / Replicate releases.
[?] how ranking works
- 501+ exceptional
- 100–500 strong
- 51–99 mid
- 30–50 fresh
- <30 quiet
- live — currently trending
- cooling — losing momentum (6–18h)
- frozen — no longer trending; ranking decays over time
Composite ranking: trend velocity + log(raw signal) − staleness penalty. The bar shows relative strength within the current batch, higher = stronger overall position.
Mistral Large 4
Mistral Large 4 (le Chonk) is a 1T-parameter, natively multimodal model with 49B active parameters, offering state-of-the-art open-weight performance in cybersecurity, coding, and agentic workflows, with weights releasing by month's end.
Mistral Large 4: "Le Chonk"
Mistral has released Mistral Large 4, a new flagship model named 'Le Chonk', available on their cloud platform.
AI is now capable of developing its own inference hardware
openTPU is an open-source AI accelerator designed and developed by AI agents, running on an FPGA card and matching simulator outputs for models like LFM2.5 and Qwen3 with 82-94% DRAM peak bandwidth.
Meta’s Muse is an adorable privacy and security dumpster fire
Meta's Muse AI agent launched with critical security flaws, including a zero-day vulnerability, unauthorized access to private messages, and a rushed codebase, despite Meta's claims of prioritizing privacy and security.
Vibecoding isn't as fun as writing code by hand
The author argues that AI-assisted 'vibecoding' front-loads the fun of building software, sacrificing the long-term satisfaction and learning that comes from writing code by hand, though it enables projects that would otherwise be impossible.
Polars 2.0
Polars 2.0 adds out-of-core support, streaming engine defaults, first-class SQL, and a new Map dtype, leading to performance wins over DuckDB and DataFusion in benchmarks.
Former German spy chief arrested for attempted treason
Former German spy chief arrested for attempted treason, sparking debate on German sympathy for Russia.
Tapo (Rust/Python library) now speaks TP-Link's TPAP protocol
The Tapo Rust/Python library v0.11.1 now supports the TPAP protocol, allowing users to control devices with the 'Third-Party Compatibility' switch off, which is now the default for newer firmware.
The Early History of Smalltalk (1993)
Alan Kay's paper details the early history of Smalltalk, tracing its origins from 1960s ARPA research and Xerox PARC innovations like overlapping windows and object-oriented design to the creation of the personal computer concept.
Mathematics of Geothermal Energy
Geothermal energy harnesses Earth's internal heat for heating and electricity, using mathematical models to optimize reservoirs and predict geological impacts like subsidence and earthquakes.
Subquadratic 3SUM and Subcubic APSP
Researchers present a new algorithm for thin matrix products that yields the first polynomial improvements over textbook algorithms for 3SUM and APSP, solving 3SUM in O(n^1.9992) time and APSP in O(n^2.9995) time.
Show HN: Parseable, an open observability datalake, handles 100M time-series/min
Parseable is an open observability datalake capable of handling 100M time-series per minute.
A list of Free Software network services and web applications which can be hosted on your own servers
A curated list of free software network services and web applications for self-hosting, categorized by function like analytics, email, and CMS.
Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥
Firecrawl is an open-source API for web scraping and data extraction, offering search, crawl, and agent features with LLM-ready output and high reliability.
The agent that grows with you
Hermes Agent is a self-improving AI agent with built-in learning loops, memory, and multi-platform support (Telegram, CLI, etc.), running on any model via Nous Portal or OpenRouter, and installable via a simple script.
Never stop coding. Free MIT AI gateway: one endpoint, 359 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by hundreds of contributors
OmniRoute aggregates 350+ AI providers and 90+ free tiers into one endpoint, offering ~1.51B free tokens monthly via RTK compression and 19 routing strategies.
Turn any idea, plan, or codebase into a beautiful interactive diagram. An agent skill for Claude Code, Codex, and more.
Archify is a Node.js tool that converts codebase JSON into interactive, shareable system architecture diagrams for AI coding agents like Cursor and Claude.
Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
Colibrì is a pure C engine running the massive 744B GLM-5.2 MoE model on consumer hardware by streaming experts from disk, featuring MLA attention, int8 MTP speculation, and 9.9GB resident RAM usage.
Spotify, native and fast. One lightweight Rust app for your whole library, local playback, and Spotify Connect on Linux, macOS, and Windows.
Spotifast is a fast, native Spotify client written in Rust that uses 100-250 MB RAM, starts in under a second, and requires Spotify Premium for playback, with features like device control, library browsing, and playlist editing.
Google's latest fast image generation model with sharp text rendering, conversational editing, multi-image fusion, and up to 4K output
Google's Nano Banana 2.1 is a fast, high-efficiency image generation model that creates photorealistic scenes, renders accurate text, and edits images with up to 14 reference inputs, supporting 1K to 4K resolutions and various aspect ratios.
A GUI for your coding agents
MonoCode is a desktop UI for coding agents that supports Claude Code, Cursor, and others, offering CLI access via /operator commands to manage sessions and folders.
RIMES — modern macOS IME (rime-scholay): librime + buffer workbench
RIMES is a multi-platform input method based on RIME, featuring Buffer, Capsule, and Mailbox for translation, clipboard history, and AI chat, with macOS, Android, and Windows support.
Make videos on the computer you already own. FreeVideo runs MiniMax H3 in as little as 8 GB of VRAM and 16 GB of RAM, and adapts its acceleration path to your hardware.
FreeVideo is a local inference engine for MiniMax H3 on consumer GPUs, running on as little as 8GB VRAM and 16GB RAM, with hardware-adaptive execution and ComfyUI integration.
EmbeddingGemma 2 was evaluated across text, code, vision, visual document, video, and audio embedding benchmarks. All results reported below use the full-precision checkpoint.
Google's EmbeddingGemma 2 is a 740M parameter multimodal model mapping text, images, video, and audio into a 768d vector space, optimized for on-device use with 8K context, 100+ languages, and Matryoshka Representation Learning for 6x storage reduction.
<!-- ### quantize_version: 2 --> <!-- ### output_tensor_quantised: 1 --> <!-- ### convert_type: hf --> <!-- ### vocab_type: --> <!-- ### tags: --> <!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K...
This is a GGUF quantized version of the TranslateGemma 12B conversational model, optimized for local inference via llama.cpp, Ollama, and other compatible tools.
kandinskylab/Kandinsky-6.0-Pro-5s-Diffusers
Kandinsky 6.0 Pro is a 29B parameter diffusion model for generating 5-second synchronized audio-video clips in T2AV and TI2AV modes, with a separate super-resolution pipeline for Full-HD output.
English · 日本語 **Baberu OCR reads the text inside a manga speech bubble.
Baberu OCR is a 115M multilingual model (Japanese, Chinese, English) that reads manga speech bubbles, beating its teacher on Japanese and matching a much larger model on Chinese and English, with v1.1 supporting up to 256 characters.
This repository contains **Kandinsky 6.0 Pro, pretrained**: the pretrained Pro checkpoint. It uses the same architecture and sampler settings as the main Pro model.
Kandinsky 6.0 Pro is a pretrained 29B parameter diffusion model for generating 5-second synchronized text-to-audio-video clips with lip-sync and Full-HD output, available via Hugging Face Diffusers.
This repository contains **Kandinsky 6.0 Lite, pretrained**: the pretrained Lite checkpoint. It uses the same architecture and sampler settings as the main Lite model.
Kandinsky 6.0 Lite is a 3B parameter diffusion model for generating 5-second synchronized text-to-audio-video clips with lip-sync, using the Kandinsky6TI2VAPipeline and Kandinsky6SRPipeline for super-resolution.
This repository contains **Kandinsky 6.0 Pro, distilled**: a distilled Pro checkpoint that samples in 10 steps with the PiFlow scheduler.
Kandinsky 6.0 Pro Distill generates 5-second synced audio-video clips in 10 steps with guidance 1.0, supports T2AV and TI2AV modes, and includes a separate super-resolution pipeline for HD output.
A NInfer **v3** artifact of ISTA-DASLab's Qwen3.
This is a 3.5-bit quantized NInfer v3 artifact of Qwen3.8-27B GSQ-RCO, optimized for RTX 3090/4090/5090, achieving perplexity of 7.07 and high benchmark scores like 100% on AIME 2025.
Interfacing Integrated Management System (IMS) — Security SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial informational page for Interfacing Integrated Management System (IMS), a Business Process Management software offering a free trial, not a direct download source.
NeuBird — Security SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This GitHub repository is a tool for downloading content from the Neubird platform, currently with 10 stars and no forks.
MOVEit — Security SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial informational repository for MOVEit, a security-focused Managed File Transfer (MFT) software by Progress, offering a free trial and available on any platform.
OfficeSpace Software — Communications SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial informational repository for OfficeSpace Software, a SaaS facility management tool, not a download source.
<!-- ### quantize_version: 2 --> <!-- ### output_tensor_quantised: 1 --> <!-- ### convert_type: hf --> <!-- ### vocab_type: --> <!-- ### tags: --> <!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K...
Darwin-9B-KOREA-GGUF is a 9B parameter Apache-2.0 licensed Korean-English bilingual model available in GGUF format for use with Transformers, llama.cpp, Ollama, and other local apps.
<!-- ### quantize_version: 2 --> <!-- ### output_tensor_quantised: 1 --> <!-- ### convert_type: hf --> <!-- ### vocab_type: --> <!-- ### tags: --> <!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K...
Darwin-28B-KR-Legal-GGUF is a Korean and English legal model available under Apache-2.0, optimized for GGUF formats and local inference via llama.cpp, Ollama, and Transformers.
1Password — Security SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial informational repository for 1Password, a security password manager, which provides links to the official site and a free trial offer.
HSI Donesafe — Security SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial informational repository for HSI Donesafe, an EHS Management SaaS offering a free trial, not a direct download source.
MaintainX — AI & Productivity SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial GitHub repository providing information and links for MaintainX, a CMMS SaaS product offering a free trial, not a direct download.
NMI Payments — Business SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial GitHub repository providing information and links for NMI Payments, a business SaaS payment processing software offering a free trial, not the software itself.
Google Workspace — Business SaaS listing. Visit link via SOFTGIT. Third-party; rights belong to original authors.
This is an unofficial GitHub repository providing informational links and screenshots for Google Workspace, a business SaaS offering a free trial, rather than hosting actual software downloads.
ESP32-C5 Wi-Fi SDR spectrum analyzer (2.4/5 GHz) using ESPARGOS/esp-sdr firmware
A PC app using the ESP32-C5's Wi-Fi 6 radio as an SDR to visualize 2.4/5 GHz spectrum, waterfall, and packets via Python (PySide6) and ESPARGOS firmware.
This repository contains **Kandinsky 6.0 Lite, distilled**: a distilled Lite checkpoint that samples in 10 steps with the PiFlow scheduler.
Kandinsky 6.0 Lite is a distilled 3B parameter model for generating 5-second synchronized audio-video clips in 10 steps with guidance 1.0, supporting T2AV and TI2AV modes.
This repository contains **Kandinsky 6.0 Lite**: the 3B checkpoint of the Lite line.
Kandinsky 6.0 Lite is a 3B parameter diffusion model for generating 5-second synchronized text-to-audio-video clips with lip-sync and Full-HD output via a separate super-resolution pipeline.
<!-- ### quantize_version: 2 --> <!-- ### output_tensor_quantised: 1 --> <!-- ### convert_type: hf --> <!-- ### vocab_type: --> <!-- ### tags: --> <!-- ### quants: x-f16 Q4_K_S Q2_K Q8_0 Q6_K...
This is a GGUF quantized version of the Gemma 27B translation model, optimized for local use with tools like llama.cpp, Ollama, and LM Studio.
Self-hosted MCP server for image generation and editing with the uncensored Qwen-Image-2.1 model
A self-hosted MCP server for generating and editing images using the uncensored Qwen-Image-2.1 model, supporting text-to-image, editing, upscaling, and seamless textures via Docker on GPU or CPU.
```python import torch from diffusers import Kandinsky6SRPipeline from diffusers.
Kandinsky 6.0 VSR is a Diffusers pipeline for video super-resolution using flow-matching, generating 5-second clips up to 1080p with 4 Euler steps.
<!-- Provide a quick summary of what the model is/does. -->
huggingface project pi05-libero by lerobot with 5 stars.
A small language model trained **from scratch**.
Queen v1.5 is a 42.1M parameter crypto-focused model trained from scratch on crawled pages, chat-tuned with 97,726 Q&A pairs generated by deepseek-flash, achieving 3x higher correctness than v1.
Signal briefs — our take on the top picks
- HNMeta’s Muse is an adorable privacy and security dumpster fire
- HNAI is now capable of developing its own inference hardware
- HNVibecoding isn't as fun as writing code by hand
- HNMistral Large 4: "Le Chonk"
- HNJetBrains reported a net financial loss first time in its tracked history
- HNMistral Large 4
- HNHigh Diesel Prices Bankrupted 16 Trucking Companies in Just 30 Days
- HNAnthropic Subscriptions Offer 5x+ More Value Than OpenAI
- HNNobel Prize in Physics goes to Francis Halzen
- HNOpen Source as We Know It Is Dead
- HNWhy Common Lisp is now the best programming language
- HNFind the flattest route between any two points in SF