GPT-5.6 Luna
ChatGPT subscription · Codex CLI
0 original challenges. 4 Counterexample attempts. Every answer on the record.
Not run.
This model has only been tested on Counterexample Lab 1.1.
4 attempts. Every result.
low reasoning · faulty versions exposed · preliminary calibration
No original-suite results
No historical coding score is assigned to this model. Its saved evidence is the Counterexample Lab attempts below.
What did its tests actually catch?
A model must predict the correct ledger’s outputs exactly before its tests can earn credit for exposing a fault. These counts show how often GPT-5.6 Luna caught each fault.
- 01Lost request history3 / 3 attempts
- 02Forgotten rejection3 / 3 attempts
- 03Unchecked payload3 / 3 attempts
- 04Revalidated retry3 / 3 attempts
- 05Partial transfer0 / 3 attempts
- 06Missing revision3 / 3 attempts
- 07Stale write2 / 3 attempts
- 08Overwritten history1 / 3 attempts
low reasoning · 12 events
low reasoning · 12 events
low reasoning · — events
low reasoning · 12 events
Same frozen contract and settings across all attempts. 8 of 8 graded traces matched the reference. All responses are preserved, including weaker attempts and unscored failures.