Model evidence

GPT-5.6 Sol

Compare GPT-5.6 Sol effort levels without collapsing them into one model score.

Current batch

GPT-5.6 Sol / Max has the highest measured score at 85/100. GPT-5.6 Sol / Low is the fastest measured configuration at 39m 5s. They support different decisions.

Selected evidence
Published
Aug 26, 2026, 11:00 PM UTC
Question pack
coding-fast-v4.10
Coverage
5 measured configurations
Batch
snapshot-2026-08-26T23-00-00Z-r3

Measured configurations

ConfigurationRankScoreElapsedReference costRoute
GPT-5.6 Sol / Max#185/10044m 58s$1.82Official login
GPT-5.6 Sol / XHigh#383/10045m 48s$1.26Official login
GPT-5.6 Sol / High#480/10045m 2s$0.96Official login
GPT-5.6 Sol / Medium#968/10043m 38s$0.66Official login
GPT-5.6 Sol / Low#1365/10039m 5s$0.44Official login

Five-question profile

QuestionGPT-5.6 Sol / MaxGPT-5.6 Sol / XHighGPT-5.6 Sol / HighGPT-5.6 Sol / MediumGPT-5.6 Sol / Low
Black-box Regression Audit14/2016/2015/2012/2013/20
Retry Planner Counterexamples20/2018/2017/2017/2014/20
CI Adversarial Audit18/2017/2018/2014/2013/20
Transaction Regression Design17/2016/2018/2013/2011/20
Cache Regression Test Design16/2016/2012/2012/2014/20

What the effort spread can tell you

This family is measured at several effort levels through the recorded official-login route. The page keeps every effort separate so a strong Max result does not get attributed to Low, or vice versa.

Use the highest score and fastest completion as two different signals. If they point to different configurations, the choice depends on your quality floor rather than the family name.