Did Codex Reset
GitHub

Elvis’s small Pi SDK test finds Jev Router cheaper and faster than fixed GPT-6 Sol

elvis

elvis tested Jev Router, the packaged typesafe/jev-router on OpenRouter, on a small support agent built with a custom harness using the Pi SDK. The same 8 cases were run against a fixed GPT-6 Sol baseline, for 32 real calls in total. Both setups got every case right.

The router cost $0.008 versus $0.018 and had a median response time of 1.5 seconds versus 1.9 seconds. elvis noted that the sample is small, and said caching for using Jev as a router is not fully solved. A routing-pattern implementation is published in Jev decisions in a Pi SDK harness.

#jev#openrouter#pi-sdk#gpt-6-sol

Best Taste-Bench model chooses the better direction 59.7% of the time

A paper from Microsoft and colleagues reports that the best model answers 59.7% of Taste-Bench questions correctly when choosing the better open direction in long engineering and research tasks. Forks decided by later evidence are much harder, a larger reasoning budget does not raise accuracy, and distilling a teacher’s outcome judgment improves held-out SWE-bench Pro success.