Summary:
- Google has unveiled Gemini 3.8 Flash, its third Flash model within a short span of six weeks, emphasizing enhanced agentic workflows.
- Gemini 3.8 Flash “exceeds expectations” in tackling complex tasks, surpassing larger models on multiple metrics while being cost-effective.
- Gemini 3.8 Flash Cyber offers advanced vulnerability detection and automated patching features, though access is restricted to Google’s Fairwind Program participants.
The previous version, Gemini 3.6 Flash, debuted in late July, and Google quickly followed up with 3.7 Flash. Now, just three weeks later, they are launching the successor to 3.7 Flash.
The latest release, Gemini 3.8 Flash, aims to provide cutting-edge intelligence for agentic workflows and cybersecurity, available in two variants.
This release marks Google’s third Flash introduction in a mere six weeks, with no new frontier models anticipated. The new models are Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, with the former serving as the company’s “most intelligent workhorse” and the latter being designated for cybersecurity needs.
Gemini 3.8 Flash is primarily targeted at developers creating agentic applications and workflows, but it is also accessible to regular users who subscribe to Google AI Pro and Ultra via the Gemini app.
Google reports that Gemini 3.8 Flash outperforms many larger frontier models in the DeepSWE v1.1 benchmark, known for evaluating autonomous problem-solving capabilities in engineering. The model achieved a score of 73.7%, trailing slightly behind Claude Opus 5 at 74%, but significantly ahead of predecessors like Gemini 3.7 Flash and others.
In addition to the above, Gemini 3.8 Flash excelled in various areas, including financial assessments, legal workflows, information synthesis, terminal coding, long video comprehension, and more.
Google highlighted that “3.8 Flash works harder,” which contributes to its improved performance. For intricate tasks, the model is capable of executing additional reasoning steps when necessary, which may require more tokens but enhances overall capability.
The pricing for Gemini 3.8 Flash starts at $0.75 per million input tokens and $3.75 per million output tokens. This is an introductory rate, set to transition to $1.50 and $7.50, respectively, but it remains competitively priced compared to alternatives such as Claude Opus 5.
As for Gemini 3.8 Flash Cyber, access is restricted and available only through the Fairwind Program, targeting select government and critical infrastructure entities to utilize the model.
The model scored 86.2% on the industry-standard benchmark designed for detecting vulnerabilities in C/C++, surpassing its predecessor and larger systems like Mythos 5. It also demonstrated robust results in real-world vulnerability assessments across various programming languages, achieving a success rate of 71%—notably higher than previous iterations.
One standout feature of Gemini 3.8 Flash Cyber is its automated patch generation capability. The model reportedly produces 2.6 times more accurate patches for vulnerabilities than much larger commercial models.
While regular users might not interact directly with these advancements, their positive implications for developers and defenders can enhance overall cybersecurity in everyday technology.
Gemini 3.8 Flash is currently being rolled out to developers, enterprise clients, and those subscribed to Google AI Pro and Ultra. As noted, Gemini 3.8 Flash Cyber is exclusively available to Fairwind Program participants.
Thank you for joining our community. Please review our Comment Policy prior to posting.
