By skyforbes I haven’t yet tested its coding capabilities but when it comes to research and technical writing, it actually feels worse, hallucinates a lot more (often refers to file line like 2628 when there are 150 lines total), falls to follow instructions (especially those in user memory, such as spelling or em dashes, etc.), and fails to produce long pieces of content at once (shortens ~2000 words to ~1200 words in a document where only additions were asked without removing anything). I personally found the update quite frustrating and there is no way to go back to Gemini 2.5 Pro within the web interface, unless using API or AI studio.