Gemini 3.8 Flash released
On 2 September 2026 Google released Gemini 3.8 Flash, which it calls the most intelligent model of its Flash line, in the Gemini API and Google AI Studio, Gemini Enterprise and, for Google AI Pro and Ultra subscribers, the Gemini app, AI Mode in Search and Sheets. The introductory price is $0.75 and $3.75 per million input and output tokens, as for 3.7 Flash. Google calls it its third Flash release in six weeks.
Why it matters
Short cycles in the cheap tier: three Flash versions in six weeks. The previous one, 3.7 Flash (13 August, a record in the atlas), was deprecated on 8 October and its requests were routed automatically to 3.8 Flash, according to the Gemini API release notes.
What Google's post says. The model gives 'significant improvements' over 3.7 Flash in software engineering, agentic tasks and multi-step reasoning in specialised fields; it 'works harder', in Google's words: it takes extra reasoning steps and calls tools iteratively, so at higher effort levels it may use more tokens. For tasks where efficiency matters most, Google suggests lower effort levels or continuing with 3.7 Flash. The price is introductory, $0.75 / $3.75 per million tokens. HLE-Verified: 54.9%. On DeepSWE v1.1 the model, Google says, outperforms most larger frontier models (the text gives no number); it also outperforms 3.7 Flash on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark (no numbers). Safeguards against misuse in chemical, biological, radiological and nuclear and cyber offence; prompt-injection robustness, as measured by Gray Swan, has improved. Date. The post is dated 2 September 2026; the same day stands in the Gemini API release notes: 'Gemini 3.8 Flash generally available (GA): Released gemini-3.8-flash'. What the record does not claim. All figures are Google's own; no independent measurements were read. The cyber variant Gemini 3.8 Flash Cyber and the Fairwind Program have a record of their own.