Daily pipeline scrapes top HN posts, uses Gemini to detect OpenRouter catalog models and score per-comment sentiment, logged to a public Google Sheet.
Key Takeaways
Pulls 200 top HN posts per 24h window, filters to ~50 LLM/coding-relevant posts, then sends titles and comments to Gemini for model detection and sentiment scoring.
Model list is sourced from the OpenRouter catalog, making results directly actionable for API-based tooling decisions.
Full audit trail is logged to Google Sheets: comment IDs, detected models, and per-comment sentiment are all inspectable.
10-day trailing aggregate drives the Top 10 Model Popularity chart, combining raw mention count with sentiment breakdown.
Hacker News Comment Review
Claude leads in raw mentions but carries net negative sentiment driven by API pricing and downtime complaints; GPT-5.5 trails in mentions but shows stronger positive feedback, complicating any simple popularity ranking.
Open-weight models Kimi 2.6, Qwen 3.6, and DeepSeek show disproportionately positive sentiment relative to mention share, while DeepSeek V4 Pro is flagged as having high hype relative to benchmark performance compared to underreported MiMo V2.5 Pro.
DeepSeek Flash’s effective input cost (~6 cents/million tokens with cache hits) is called out as a better efficiency story than V4 Pro, though commenters debate whether cost matters much when heavy models dominate real coding workloads.
Notable Comments
@cheesecakegood: DeepSeek Flash at 6 cents/million cached input tokens is the normal price, not a promo – argues it outperforms V4 Pro on cost efficiency.
@jatins: Flags that Gemini appears absent from HN discussions despite being the model powering this pipeline.
