Gemini 3.8 Flash currently leads the DeepSWE leaderboard, a benchmark that evaluates how well AI models solve complex software engineering tasks. At current discounted rates, Google’s model also delivers this performance at a lower cost. Google reportedly delayed Gemini 3.5 Pro after its coding abilities failed to match competing models. However, if these benchmark results are accurate, Gemini 3.8 Flash is now challenging the leading AI coding models.
Computer-use tasks remain a challenge for Google’s AI models. Although Gemini 3.8 Flash improves on Gemini 3.7 Flash in the OSWorld-2.0 benchmark for operating-system and computer-use agents, it still trails the market-leading Claude Opus by a significant margin. OpenAI’s GPT models also perform only modestly better in this test.
Most users will not interact directly with Gemini 3.8 Flash Cyber, which replaces the previous 3.5 version. However, specialized AI models like this are becoming increasingly valuable for cybersecurity and other technical applications. Google says internal testing shows that Gemini 3.8 Flash Cyber is a major improvement over earlier versions, identifying more vulnerabilities and producing working patches more often.
According to Google, the Chrome security team found that Gemini 3.8 Flash Cyber improved patch accuracy by 2.6 times. Google’s cloud team also reported that the model discovered critical vulnerabilities within two hours. Security companies including Wiz and Palo Alto Networks have issued statements supporting the capabilities of Gemini 3.8 Flash Cyber.
Gemini 3.8 Flash is available across Google’s ecosystem starting today. Gemini 3.8 Flash Cyber, meanwhile, remains limited to trusted testers and government users. As with previous Flash releases, access to Gemini 3.8 Flash in the Gemini app requires a Pro or Ultra subscription. Users who want to try the model for free can experiment with it through Google AI Studio.
Source: arstechnica.com