Could Gemini 3.5 Flash have 250 billion to 300 billion total parameters? A Hacker News user estimates from TPU performance

Hacker News users estimated Gemini 3.5 Flash at roughly 250B–300B total parameters, with about 10B–16B active, based on TPU 8i constraints.

The model’s pricing drew attention because it appears much higher than a typical Flash-tier product, especially for token-heavy workloads.

Benchmarks and outside measurements suggest strong performance, but real-world behavior is mixed, with some users seeing overelaborate or inconsistent outputs.

The discussion implies Google’s fast model may be larger and costlier than expected, which matters for deployment, coding, and agent use.