Kamfanin Google ya gabatar da sabon model din AI mai suna Gemini 2.0, wanda aka samar don agentic experiences. Model din ya kunshi Gemini 2.0 Flash, wanda shine model din aiki da ƙarancin latency da ingantaccen aiki. Demis Hassabis, CEO na Google DeepMind, da Koray Kavukcuoglu, CTO na Google DeepMind, sun bayyana cewa Gemini 2.0 Flash ya gina ne a kan nasarar 1.5 Flash, model din da ya fi shahara a yanzu ga masu ci gaban software.
Gemini 2.0 Flash ya fi yawa a wasu manyan benchmarks, inda ta kai 92.9% a cikin ayyukan code generation a Natural2Code, wanda ya fi 85.4% na Gemini 1.5 Pro. A fannin hujja na lissafi, 2.0 Flash ya ci 89.7% a MATH da 63% a HiddenMath, wanda ya fi yawa da na gabata. A kan factuality, 2.0 Flash ya ci 83.6%, lissafin ingantaccen amsa. Model din ya kuma nuna ingantaccen aiki a cikin ayyukan multimodal, inda ta ci 70.7% a MMMU.
Gemini 2.0 Flash zai samu a matsayin model din gwaji ga masu ci gaban software ta hanyar Gemini API a Google AI Studio da Vertex AI. Kamfanin ya bayyana cewa samunwa gama gari zai biyo baya a watan Janairu tare da sauran saukar model. Google kuma ta sanar da sabon Multimodal Live API, wanda zai baiwa masu ci gaban software damar gina applications dinamiki da masu aiki.
Project Astra, wanda aka gina da Gemini 2.0, zai aiki a matsayin universal AI assistant, wanda zai iya yin magana, haÉ—a kayan aiki kamar Google Search, Maps, Lens, da sauransu. Project Mariner kuma zai bincika ayyukan agentic a cikin browsers, kamar yadda ake biyan ayyuka ta hanyar text, code, da hotuna.
