Hacker News new | ask | show | jobs
by veunes 200 days ago
I'm not surprised. Roleplay means endless sessions with huge context (character history, world, previous dialogues). On commercial APIs (OpenAI/Anthropic), that long-context costs a fortune. On OpenRouter, many OSS models, especially via providers like DeepInfra or Fireworks, cost pennies or are even free, like some Free-tier models. The RP community is very price-sensitive, so they massively migrate to cheap OSS models via aggregators. It skews the stats but highlights a real niche for cheap inference