Gemini 3.5 FLASH: BAD to OUTSTANDING

The newly available Gemini 3.5 FLASH AI model tested on my own causal REASONING TEST. Since I test this reasoning model on "HIGH", the end result is excellent. But analyzing the reasoning trace of Gemini 3.5 FLASH in details shows, any exit before 5 minutes pure runtime (on Google TPU) would have delivered a "bad" (10 step solution) qualitative result. Multiple other tests with "MEDIUM" reasoning (not recorded) confirm this feeling. Note: This test is not a statistically significant event and only present a single test for a very limited number of repetitions. Therefore the result of my test is just for a first feeling, regarding the new GEMINI 3.5 FLASH. #airesearch #aiexplained #geminiai #googledeepmind