GEMMA 4 AI Models vs. GPT-5.4

In a video titled “NEW GEMMA 4 beats GPT-5.4: The A4B Model,” released by Discover AI on April 3, 2026, an evaluation of Google’s new open AI models was conducted, focusing on GEMMA 4. The dense GEMMA 4 31B model and the GEMMA 4 MoE 26B-A4B model were rigorously tested for their capabilities in scientific reasoning, logic, and causal deduction. The findings were astonishing, as the GEMMA 4, particularly the tiny 4B model, demonstrated superior cognitive flexibility compared to larger models such as the GPT-5.4, which unexpectedly couldn’t produce a viable sequence in the test scenario. The GEMMA 4’s architecture showed promise, potentially redefined expectations for smaller models, especially in logic and strategic testing environments.