Google Announces Gemini 3 Pro: Higher Scores Than Previous Models, but Prices Increase

It is explained that this model inherits the capabilities of Gemini 2.5 Pro while improving overall performance. The knowledge cutoff is January 2025, supporting 1 million input tokens and up to 64,000 output tokens.

Multimodal input for text, images, audio, and video is also possible. According to the model card provided by Google, the model recorded scores exceeding those of Claude 4.5 Sonnet and GPT-5.1 in major benchmarks.

While independent verification is still awaited, the author stated that they have no doubts regarding the reliability of the figures. The pricing will be higher than Gemini 2.5 Pro, but remains cheaper than Claude Sonnet 4.5.

Based on preview access in AI Studio, the author attempted alternative text generation from images of benchmark tables. The model output descriptions including detailed numerical values for screen-reader users.

Furthermore, a test was conducted to process the audio of a 3-hour and 33-minute city council meeting using speech recognition. A 74MB m4a file was passed to the model with a request for transcription.