Gemini 3.8 Flash

Google's September 2, 2026 Flash — its third in 43 days — at identical specifications and identical introductory pricing to 3.7 Flash ($0.75/$3.75 per Mtok until December 31, 2026, then $1.50/$7.50), and ahead of 3.7 Flash on every published row, so there is no reason to stay on the older model. Google's pitch is that it 'works harder': more reasoning steps and iterative tool calls on complex tasks. Available in AI Studio, the Gemini API, Antigravity, Gemini Enterprise and the Gemini app for AI Pro/Ultra subscribers. The honest reading of Google's own 14-row table is that it wins 8 and loses 5 to Claude Opus 5, and the losses are where the work is open-ended: Terminal-Bench 4.0 19.1 vs 51.8, OSWorld-2.0 59.0 vs 75.4, GDPval-AA v2 1545 vs 1824. It is excellent at bounded professional tasks — finance, law, charts, video, biology — at roughly a seventh of Opus 5's price, and its DeepSWE v1.1 73.7 is within noise of Opus 5's 74.0. Widely-circulated DeepSWE figures of 74% for this model come from the public Datacurve leaderboard rather than Google's table. Its FrontierCode 1.1 rows are deliberately omitted here: the only published values come from a competitor's comparison table and are identical to 3.7 Flash's, which cannot be distinguished from a carried-forward figure.