Gemini 3.2 Flash Leaked Before I/O: Sub-200ms Inference, 92% of GPT-5.5 Performance at 1/15th the Cost

NPowerUser·high signal

Google's Gemini 3.2 Flash appeared silently in the Gemini iOS app and AI Studio on May 5, then showed up on LM Arena benchmarks. Early results show 92% of GPT-5.5 coding/reasoning performance with sub-200ms latency and 15-20x cost reduction. Outperforms Gemini 3.1 Pro on certain creative coding benchmarks. Expected official announcement at Google I/O May 19-20.