GPT-5.6 Sol/Luna/Terra benchmarked by role — Sol high wins main-session slot
A small role-based benchmark compared GPT-5.6's Sol, Luna, and Terra variants across three strategic decision memos, one -grounded execution brief, and two bug-fix-with-tests tasks. Scoring covered judgment, , risk awareness, boundary-setting, actionability, and efficiency, and the evaluator could see which model produced each answer, so it was not a blind test.
On the multi-layer productization decision, the Fable 5 scored 95, Sol max scored 94, Sol xhigh 90, Sol high 87, and GPT-5.5 high 81. On a decision-method review, the scored 92 with Sol max close behind at 91.
On a bounded-opportunity comparison, Sol high came within one point of the (94 vs 95) while using only 81.5 seconds and 369 ; Sol max took 216.9 seconds and 5,178 without changing the actual decision. On the -grounded execution brief, Sol high scored 93 (80.73s, 1,818 tokens) and Sol medium scored 91 (70.66s, 779 tokens) — medium was about 12.5% faster and used 57% fewer while nearly matching high's quality.
Key points
- Compared GPT-5.6's Sol/ across 3 strategic memos, 1 repo-grounded brief, and 2 bug-fix tasks
- Sol high matched the within 1 point (94) using just 81.5s and 369 , earning the main-session pick
- Sol max spent 5,178 (216.9s) yet reached the same decision as Sol high
- On the execution brief, Sol medium used 57% fewer than Sol high while scoring only 2 points lower
- was non-blind (model identity visible) with a small sample, so results are directional, not conclusive