Archestra Apps HackathonSee what people are building

Thread

SM
Sakshi M4:17 PMOpen in Slack
Running into repeated LLM provider issues while building an app in
Archestra — tried OpenRouter (empty responses, then a 402 error saying
it requested 32,768 max_tokens but my free account can only afford
1,600-4,000 depending on the moment), Gemini free tier (rate-limited
after ~5 requests), and OpenAI (empty response with no clear reason).
The OpenRouter 32k max_tokens default seems like it might be worth
lowering for free-tier accounts, or exposing as an easy setting — right
now it silently fails instead of scaling down. Is there a way to
configure max_tokens per-model, or a recommended provider/model combo
that works reliably on free tiers for iterative app building?

1 reply
A
Archestra4:17 PMOpen in Slack
Good feedback on the 32k default — that's definitely too high for free tiers. I'll check if there's a way to configure maxtokens per-model. Based on what I know: there's currently no per-model `maxtokens config exposed in the UI or easy settings. The 32k default is baked in and causes exactly the issue you hit with free-tier OpenRouter accounts. **For now, workarounds:** - **Groq** free tier tends to be more reliable for iterative building (generous limits, fast) - **OpenRouter** with a model that has lower default max_tokens (like mistralai/mistral-7b-instruct) may work better - Some users have had luck with **Anthropic** free credits if you have them I'd suggest raising this as a feature request in the hackathon feedback — exposing max_tokens` as a configurable setting (or auto-scaling based on account limits) would help a lot. The silent failure is definitely a bug worth flagging. 🤖 Slack Agent