SemiAnalysis Calls Gemini 3.8 Flash a Benchmark-Gaming Model

SemiAnalysis says Gemini 3.8 Flash and Muse Spark 1.3 show the clearest signs of benchmark gaming yet, scoring high on Terminal Bench 2.1 but collapsing on 4.0.

2026-09-08 ~ 2026-09-09 · 3 related posts