Google Gemini 3.8 Flash Official Release EvaluationProvider report
2026-09-02Based on Google's official September 2, 2026 announcement, the complete DeepMind model-card performance table, and its methodology. The source table contains six models and 13 benchmark rows. This site splits the compound LVBench agentic / static and BioMysteryBench Human Solvable / Human Difficult rows, producing 15 metrics and 85 available scores. Only Gemini 3.8 Flash has a published LVBench agentic value; blank cells do not create scores. Gemini scores use the Gemini API (model-id gemini-3.8-flash), default sampling settings, and pass@1 unless otherwise noted; DeepSWE 1.1 uses high thinking. Full performance table: https://deepmind.google/models/model-cards/gemini-3-8-flash/ Methodology: https://deepmind.google/models/evals-methodology/gemini-3-8-flash