HN SOTA: Tracking Coding Model Popularity via Hacker News Comments
HN SOTA is a data pipeline that automatically tracks which coding models are popular on Hacker News. It runs daily, scraping the 200 most popular posts from the past 24 hours, then prompts Gemini to filter those whose titles are about LLMs or coding in general (max 50). For each relevant post, it sends the title and comments to Gemini, asking it to identify models from the OpenRouter model list and rate sentiment (positive/negative/neutral) per comment per model.
Results are logged to a public Google Sheet, including comment IDs so you can audit by appending the ID to https://news.ycombinator.com/item?id=. The main page shows a top 10 leaderboard of model mentions over the trailing 10 days, with bar charts scaling to 100%.
The pipeline is designed for transparency – the author explicitly wanted the ability to sanity-check model outputs. Future iterations may scan for harnesses (e.g., Aider, Continue) and hardware setups.
As of May 1, 2026, the aggregate covers April 23 to May 1. The Google Sheet linked from the site gives granular per-comment data.
If you follow model discussions on HN but can't read every thread, this tool gives a quick, automated summary of what's being talked about and how.
📖 Read the full source: HN AI Agents
👀 See Also
Google PM Open-Sources Always On Memory Agent with SQLite Storage, No Vector DB
Google senior AI product manager Shubham Saboo has open-sourced an Always On Memory Agent that stores structured memories in SQLite instead of using vector databases, running on Gemini 3.1 Flash-Lite with scheduled memory consolidation every 30 minutes.
Lumyr: Dashboard Generation via Claude with Python and Streamlit Automation
Lumyr is a tool that generates live, shareable dashboards from plain English descriptions using Claude for dashboard generation and automating the Python and Streamlit layer. Users don't need to write Python, open Streamlit, deploy, set up hosting, or manage infrastructure.
VibeIndex.ai: Searchable Hub for 90K+ AI Skills, MCPs, and Plugins with Security Scanning
A Korean AI researcher has built vibeindex.ai, a searchable hub that indexes over 90,000 AI skills, MCP servers, and plugins with hourly updates and security scanning using Cisco Skill Scanner across 17 threat categories.
Dual DGX Sparks vs Mac Studio M3 Ultra: Practical Comparison for Running Qwen3.5 397B Locally
A developer compared running Qwen3.5 397B locally on a $10K Mac Studio M3 Ultra 512GB and a $10K dual DGX Spark setup. The Mac Studio achieved 30-40 tok/s with 800 GB/s bandwidth but slow prefill, while the Sparks delivered 27-28 tok/s with faster compute but complex setup.