Google Releases Gemini 3.8 Flash, Its Third Flash Model in Six Weeks

Google has released Gemini 3.8 Flash, marking the company's third Flash model launch in just six weeks. The model currently sits at the top of the DeepSWE leaderboard, which evaluates an AI system's ability to handle complex software engineering tasks, and it reportedly does so at a lower cost under current discounted pricing.

Google had previously delayed the release of Gemini 3.5 Pro after its coding performance failed to match competitors. If the new benchmark results hold up, even Google's budget-tier Flash models are now competitive with leading models from other companies.

Computer-use skills still lag behind

Agentic computer use has remained a weak point for Google's models. Although Gemini 3.8 Flash shows improvement over the 3.7 Flash version on the OSWorld-2.0 benchmark, which tests an AI agent's ability to operate a computer, it still trails well behind market leader Claude Opus. Notably, GPT models also perform poorly on this same test.

A dedicated model for cybersecurity

Google simultaneously introduced Gemini 3.8 Flash Cyber, replacing the previous 3.5 version. Built for specialized cybersecurity work, the model reportedly showed significant gains over its predecessor in internal testing, identifying more vulnerabilities and producing working patches more consistently.

Chrome's security team reportedly saw a 2.6-times increase in patch accuracy after switching to the new model. Google's Cloud team said the model detected a critical vulnerability in just two hours. Partner companies including Wiz and Palo Alto Networks have also vouched for the new cybersecurity model's capabilities.

Gemini 3.8 Flash becomes available across Google's ecosystem starting today, though a Pro or Ultra subscription is required to use it within the Gemini app; it can also be tried for free through AI Studio. Gemini 3.8 Flash Cyber, meanwhile, remains restricted to trusted testers and government partners for now.