Developers find persisted CoT via Responses API makes low-effort Astra smarter

Developers report that because the Responses API persists chains of thought, running Astra at low or medium reasoning effort in /goal scenarios can actually perform better, and suggest letting the model dispatch Luna subagents to cut costs.

2026-09-05 ~ 2026-09-06 · 3 related posts

• Tip: persisted CoTs in Responses API may make low-effort Astra smarter in /goal — realy0usaf · 2026-09-05
• "Use Luna subagents to save costs": practical Astra orchestration tips from early users — brandon_galang · 2026-09-05
• Responses API persists CoTs, so lower effort modes may actually be smarter — aidan_mclau · 2026-09-06