Google launches Gemini 3.8 Flash and keeps Cyber for defenders

[{"h":"CONFIRMED","p":"Google stopped treating Gemini 3.8 Flash as a quota ID. On September 2 the company published the model and its sibling 3.8 Flash Cyber: a coding and agent workhorse at 3.7 prices, and a defense variant only for people inside the Fairwind Program. The morning was grayscale in the app and a slug in Cloud. The afternoon became a product sheet, a price, and an endpoint."},{"h":"The Flash cadence became a product","p":"Google's cheap line now ships on a factory clock. 3.6 Flash landed in July. 3.7 Flash arrived on August 13. Three weeks later came 3.8. The Wall Street Journal had described the model, internally Skimaki, as a coding piece ready as soon as Wednesday. The Gemini app in India and Brazil was already naming itself 3.8 while the picker stayed on 3.7. Then the identifier gemini-3.8-flash showed up on Cloud quota and usage dashboards. The company blog closes that line. The stable ID is gemini-3.8-flash, with a 1 million token context window and up to 65 thousand output tokens, and thinking levels of low, medium, and high. Introductory pricing copies 3.7: 0.75 dollars in and 3.75 dollars out per million tokens through December 31, 2026. On January 1, 2027 the rate rises to 1.50 and 7.50. It is live for Pro and Ultra subscribers in the Gemini app, AI Mode in Search, Sheets, Antigravity, AI Studio, the Gemini API, and Gemini Enterprise."},{"h":"The production pin flips","p":"Until midday the correct reading was to keep 3.7 in production and treat 3.8 as a shadow. Google now says 3.8 is generally available and becomes the default for managed agents in Antigravity. The house thesis is cheap and aggressive: the model works harder, calls tools again, and spends extra tokens at high effort to get closer to frontier quality. Artificial Analysis scored 59 on the Intelligence Index at high, three points above 3.7, at the same introductory price, with a higher cost per task because the model talks more. DeepMind sells DeepSWE, finance, and legal as proof of long-horizon agents. Those are its numbers. Independently, the AA index puts 3.8 Flash on the same reasoning floor as GPT-5.6 Sol and Grok 4.6, at a fraction of Opus pricing. Teams that need predictable latency and a closed bill stay on 3.7. Teams that want the new workhorse should test 3.8 at medium before turning high loose on volume."},{"h":"Cyber is not everyone's API","p":"3.8 is not the late flagship. It is the Flash model Google can iterate while 3.5 Pro still has no date. The Cyber variant does not go to the public API. It enters Fairwind, with the CodeMender harness, for governments, critical infrastructure, and maintainers. Google says it trained the shared core in the cyber domain and that Cyber prioritizes vulnerability fixing, not exploitation. Without program access, that is a statement, not a tool. Treating 3.8 as a win over Anthropic or OpenAI is still lab marketing. Treating the blog as a footnote after the Cloud ID would be the opposite error: the official announcement arrived, and it confirms what the infrastructure already showed this morning."},{"h":"MaxAssistant reading","p":"The Flash series is Google's cadence weapon. The official announcement lagged the quota ID, exactly as this desk described in the morning. For small companies the move now is the reverse of grayscale: a test pin on 3.8, a production pin on 3.7 only where tokens and latency still rule, and no expectation that Cyber becomes an open endpoint."},{"h":"Update","p":"September 2, 2026, 13:00 BRT: Google published Gemini 3.8 Flash and 3.8 Flash Cyber. The label rises from STRONG SIGNAL to CONFIRMED."},{"h":"Sources","p":"Google, Introducing Gemini 3.8 Flash and 3.8 Flash Cyber: https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/ | Google, Gemini 3.8 Flash API docs: https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash | Google, Fairwind Program: https://blog.google/innovation-and-ai/technology/safety-security/fairwind-program/ | Artificial Analysis, Google has released Gemini 3.8 Flash: https://artificialanalysis.ai/articles/gemini-3-8-flash | Google DeepMind, Gemini 3.8 Flash: https://deepmind.google/models/gemini/flash/"}]