🧠Google Releases Gemma 4 Open Model, Outperforms Llama 4 Maverick
Gemma 4 Beats Llama 4: Google Raises the Open Model Bar
TL;DR
Google's Gemma 4 31B outperforms Llama 4 Maverick on AIME 2026, GPQA Diamond, and LiveCodeBench benchmarks across math, reasoning, and coding.
Google announced Gemma 4, the latest state-of-the-art open model that serves as the foundation for the next generation of Gemini Nano. The 31B parameter model outperforms Meta's Llama 4 Maverick across key benchmarks: 89.2 percent vs 88.3 percent on AIME 2026 for math, 84.3 percent vs 82.3 percent on GPQA Diamond for reasoning, and 80.0 percent vs 77.1 percent on LiveCodeBench for coding. The release reinforces the shrinking gap between open-source and proprietary AI models.
Key Points
Google released Gemma 4, a new open model outperforming Llama 4 Maverick on math, reasoning, and coding.
Gemma 4 31B scored 89.2% on AIME 2026 vs Llama 4 Maverick's 88.3%.
On LiveCodeBench, Gemma 4 scored 80.0% compared to Llama 4 Maverick's 77.1%.
Gemma 4 serves as the foundation for the next generation of Gemini Nano on Android.
Why It Matters
Gemma 4's benchmark-beating performance signals that open-source AI is reaching parity with proprietary models, giving developers powerful on-device AI capabilities without cloud dependency.
Frequently Asked Questions
Why does this matter?
Gemma 4's benchmark-beating performance signals that open-source AI is reaching parity with proprietary models, giving developers powerful on-device AI capabilities without cloud dependency.
What happened?
Google's Gemma 4 31B outperforms Llama 4 Maverick on AIME 2026, GPQA Diamond, and LiveCodeBench benchmarks across math, reasoning, and coding.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,303 builders reading daily.