Marina Lyubimova
⬤ Google has released Gemini 3 Flash, the latest member of the Gemini 3 series and early figures look strong. The model outscores Gemini 3 Pro on SWE Bench Verified plus turns in high marks on Humanity's Last Exam besides MMMU-Pro. The company is plainly focused on AI that reasons more effectively and processes multiple kinds of input together.
⬤ On Humanity's Last Exam, Gemini 3 Flash ranks first when tools are active reaching roughly 45.8 %. With tools disabled it still exceeds 43 % placing it in front of every rival tested - Gemini 3 Pro, the 2.5 family, Claude Sonnet 4.5, GPT-5.2 but also Grok 4 Fast. The outcome indicates that the system is improving at step-by-step problems that demand orderly thought.
⬤ Multimodal performance is also encouraging. MMMU-Pro, which measures how models merge text and images, awards Gemini 3 Flash about 81.2 %. That number tops Gemini 3 Pro as well as leaves earlier Gemini versions plus outside competitors behind. The outcome is further evidence that Google is gaining ground on systems which interpret varied inputs side by side.
⬤ Observers tracking the AI race will find the numbers significant. When third party benchmarks repeatedly show gains, they support Google's rapid update approach across the Gemini line. As major firms continue to stress reasoning skill and multimodal strength, progress of this sort is likely to shape market opinions on adoption speed, platform rivalry or the broader direction of AI innovation.
Marina Lyubimova
