Google unveils Gemini 3.8 Flash, a new model focused on reasoning and coding
Google announced the launch of Gemini 3.8 Flash on September 2, a new AI model designed for agentic workflows, coding, and advanced reasoning tasks. This model, a successor to Gemini 3.7 Flash, aims to advance reasoning and programming while maintaining the speed and cost efficiency of the "Flash" series. Google states that Gemini 3.8 Flash can approach the performance of more expensive, state-of-the-art models in certain evaluations. A key feature is its ability to handle complex programming tasks over long sequences, reportedly surpassing several models in autonomous end-to-end engineering problem-solving on the DeepSWE v1.1 benchmark. The model also shows improved performance in professional fields requiring analysis and reporting, such as the financial and legal sectors. Google highlights Gemini 3.8 Flash's capability to create interactive applications and experiences from simple instructions, citing examples like a 3D game and a functional DOS version of Google Maps. Additionally, Google introduced Gemini 3.8 Flash Cyber, a specialized version for cybersecurity professionals, currently accessible to a limited group through the Fairwind Program. Google claims Gemini 3.8 Flash Cyber achieves "frontier" level performance in autonomous vulnerability detection, outperforming its predecessor Gemini 3.5 Flash Cyber and larger state-of-the-art models on the CyberGym benchmark. It also boasts a success rate exceeding 70% in detecting vulnerabilities across complex codebases in 20