Google on the 2nd (local time) released "Gemini 3.8 Flash," which boosts reasoning, coding, and artificial intelligence (AI) agent performance, and "Gemini 3.8 Flash Cyber," specialized for cybersecurity tasks. It unveiled a follow-up lightweight model just three weeks after introducing "Gemini 3.7 Flash."
Google said on the 3rd on its official blog that "Gemini 3.8 Flash" is the flagship with the highest intelligence among the Flash series models introduced so far. It noted in particular that it improved multistep reasoning in software engineering, AI agent tasks, and specialized domains over the previous Gemini 3.7 Flash.
On July 21 it rolled out "Gemini 3.6 Flash," and on the 13th of last month "Gemini 3.7 Flash." With the debut of "Gemini 3.8 Flash" this time, the third Flash model in the past six weeks has been unveiled.
Google emphasized that despite being a lightweight model, "Gemini 3.8 Flash" outperformed OpenAI's top-end model "GPT-5.6 Sol" and Anthropic's "Claude Opus 5" on the long-horizon software engineering benchmark (DeepSWE v1.1), which evaluates the ability to autonomously solve complex software engineering problems end to end.
It also said it surpassed not only the previous "Gemini 3.7 Flash" but other frontier models in financial and legal analysis.
Introductory pricing is $0.75 per 1 million input tokens and $3.75 per 1 million output tokens, the same as the previous version.
Released alongside it, "Gemini 3.8 Flash Cyber" is a model strong in vulnerability detection and automatic patching. Google claimed that in Chrome browser vulnerability patch tests, "Gemini 3.8 Flash Cyber" generated 2.6 times more correct patch code than the best existing commercial model. It added that it completed a Google Cloud vulnerability detection task that previously took months in under two hours.
However, "Gemini 3.8 Flash" showed lower performance than Meta's new model "Muse Spark 1.3," which was unveiled the same day. According to AI performance analytics firm Artificial Analysis, "Muse Spark 1.3" scored 62 on the Artificial Analysis Intelligence Index (AAII), tying for third with Anthropic's "Claude Fable 5," while "Gemini 3.8 Flash" scored a lower 59.
While Google is accelerating the rollout of lightweight models, the next-generation flagship AI model "Gemini 3.5 Pro," which was originally slated for release in June, remains delayed. Some observers are even suggesting Google might skip 3.5 Pro and go straight to "Gemini 4.0 Pro."