Google has sped things up again. Gemini 3.8 Flash is here, reportedly thinking more thoroughly on complex tasks

Keeping up with the pace at which Google churns out new language models is a challenge even for tech enthusiasts. The cheaper and faster Flash series is now getting another upgrade barely a month after the previous one – Gemini 3.8 Flash is designed to be smarter, especially in code and agentic tasks, and at the exact same price as the 3.7 Flash model.

Gemini 3.8 Flash excels primarily in programming

Google positions Gemini 3.8 Flash as its smartest all-around model yet for coding and so-called agentic tasks – situations where the model plans its own process and calls various tools to complete it. The previous 3.7 Flash version was introduced only in mid-August.

The progress is expected to be most evident in long programming assignments. In the DeepSWE v1.1 test, which measures how a model solves a real software bug from start to finish, 3.8 Flash reportedly outperforms even significantly larger and more expensive competitor models. In the demanding academic HLE-Verified test, Google claims a success rate of 54.9%. The new version is also said to be better in more specialized tasks in finance and law.

Google claims that for complex tasks, the model takes extra steps, thinks more thoroughly, and uses tools more often – but this also means it consumes more tokens, so it can actually end up being more expensive, even if the rate per token remains the same.

The rate remains at $0.75 per million input tokens and $3.75 per million output tokens, exactly where 3.7 Flash started. Those who want to save can reduce the model’s “effort” with a switch, or simply stick with the older 3.7, which Google continues to fully support for economical deployment.

Second model targets cybersecurity

Alongside the standard model, Google also introduced Gemini 3.8 Flash Cyber, tuned for defense against cyberattacks. It can reportedly find vulnerabilities in foreign code and immediately suggest patches for them, with the company emphasizing that it focused on fixing flaws, not exploiting them. However, not just anyone will get the model – Google is making it available only to vetted “defenders,” typically government agencies or critical infrastructure operators, through its new Fairwind program.

The company claims that it is already deploying the model to secure its own code. The Chrome security team reportedly created 2.6 times more correct patches with it than with the best commercial models, which are significantly larger. However, these are Google’s own figures, which cannot be independently verified externally. According to the company, the new generation is also better at resisting so-called prompt injection, which is an attempt to subtly insert malicious instructions into the model.

Available now in the Gemini app, but only for subscribers

For developers, the model is available almost immediately – via the Gemini API in Google AI Studio, Android Studio, and the Antigravity environment. For Google AI Pro and Google AI Ultra subscribers, it is also immediately available in the Gemini app. It is also expected to appear in AI mode in Google Search and Google Sheets.

Do you find the pace at which Google churns out new models more appealing, or are you starting to get lost in it?

Source: Google