devwithjevAdaptive Reasoning Effort in Codex
vechen made Jev change GPT-6’s reasoning effort inside Codex during a task. In their tests, this reduced Astra costs by 50%.
View on X cost50% lower Astra costs
vechen@miu21590𝕏
People use Jev to pick a model before a task. I made it change GPT-6's reasoning effort inside Codex DURING the task. More thinking when stuck. Less for routine steps. 50% lower Astra costs in my tests. Faster runs, without breaking prompt caching. https://t.co/NrpGFfWW9v
Jev uses more reasoning when the task is stuck and less for routine steps. vechen says the approach preserves prompt caching and makes runs faster.
Also filed under Tools & apps
- Jev Auth Event Auditing and Alert Telemetry
It provides shadow-mode Jev authn/authz event auditing and alert telemetry for OAuth2 and IAM.
- Ask Questions About Git History
HekaJev lets users ask questions about Git history in plain English. The open-source tool is built using Jev.
- Local Jev-Compatible Decision-Model Runner
ollaya is a local decision-model runner with a Jev-compatible System One API.
- Open Jev-Style Decision Model
JevAny is an open Jev-style model and training stack for typed decisions and option probabilities.