Gemini 3.8 Flash – Is Google back in the game?

Where does Gemini 3.8 Flash truly outperform GPT-5.6, Claude Opus 5, Kimi K3, and GLM-5.3, where does it merely close the gap, and where are there still no grounds to speak of dominance?

Information status: September 5, 2026 • Analysis based on publicly available results and documentation.

Benchmarks from different organizations are not fully comparable.

In short: Gemini 3.8 Flash is one of the most interesting "workhorse" class models available in September 2026.

Its greatest advantage is not a single benchmark record, but the combination of reasoning quality, agentic coding, multimodality, a million context tokens, and a low input price.

However, this does not mean that Google has regained its position as the leader of the entire market.

GPT-5.6 still holds a very strong advantage in token efficiency and the agent ecosystem, Claude Opus 5 remains an extremely strong competitor in long tasks, and Chinese models Kimi K3 and GLM-5.3 show that competition does not end in the USA.

Table of Contents What did Gemini 3.8 Flash actually show?