Google releases Gemini3

Google releases three new models including Gemini3.6Flash, and postpones the release of its flagship 3.5Pro. Google DeepMind officially released three new artificial intelligence models: Gemini3.6Flash, 3.5Flash-Lite and 3.5Flash Cyber ​​on July 21 to comprehensively enhance the efficiency and application capabilities of medium and lightweight models. As the core of this update, the main model Gemini3.6Flash has achieved significant improvements in code writing, knowledge understanding and multi-modal capabilities, while reducing Token usage by up to 17%, further reducing call costs. Gemini3.5Flash-Lite is committed to creating a cost-effective and lightweight experience; while 3.5Flash Cyber​​​ is a dedicated model for network security and is open to government agencies and trusted partners in the form of a limited access pilot for vulnerability identification and repair. Google said this release is aimed at providing high-reliability, low-latency infrastructure support for enterprise customers building AI Agents at scale. It is worth noting that this update does not include the long-awaited flagship model Gemini3.5Pro. Since the model was updated in February, its release has been delayed due to internal performance compliance challenges. It is currently in the partner phase of testing, and officials said they will strive to launch it as soon as possible. Faced with fierce competition from OpenAI’s release of GPT-5.5/5.6 and Anthropic’s release of Claude Opus4.8 and Sonnet5, Google is responding to computing power and efficiency challenges through the collaboration of hardware and software. Not only has it been revealed that it has internally developed an efficient server chip code-named “Frozen v2”, but it has also launched the largest Gemini4 cutting-edge pre-training to date, demonstrating its capabilities in AI Continuous investment in infrastructure and next-generation large models. via AI News (author: AI Base)