Anybody use frontier models like Astra/Fable for planning/judging, and qwen3.8 as the main workhorse? Curious to hear about your setups!
A user is exploring a hybrid AI setup, leveraging a ChatGPT Plus subscription with Astra for high-level planning and judging, while using a local qwen3.8-27b model for coding implementation. The strategy aims to minimize cloud usage costs by having Astra plan, qwen implement, Astra critique, and qwen fix, thereby retaining high-parameter intelligence for architectural decisions. The user is seeking advice on similar setups, workflow switching, and optimal settings.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 12, 2026, 12:13 UTC
IngestedOffset at this time: UTC+0Sep 12, 2026, 22:01 UTC
- Published
- Sep 12, 2026, 12:13
- Ingested
- Sep 12, 2026, 22:01
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
Hey everyone!
I'm curious to hear from people that use a combination of cloud-based frontier models and local ones for development. I'm planning to set something similar up and wanted to hear about actual examples of this in action.
Currently my plan is to use my chatgpt plus subscription purely for planning and judging with Astra, and then run a local qwen3.8-27b model for the actual coding gruntwork - i.e Astra plans -> qwen implements -> Astra critiques the implementation -> qwen fixes and so on. This way I keep cloud usage down and cheap, while retaining the high-parameter intelligence for architecture decisions and optimization.
For those of you who have a similar setup, how is it? How do you switch between the two, what harness/settings/etc? Anything you would suggest?