Samsung just dropped fresh MLPerf numbers on the Exynos 2600, and the result is quietly impressive. While everyone obsesses over flagship Snapdragon wins, Samsung’s in-house 2nm chip (already shipping in Galaxy S26 and S26+) is showing real progress on the one thing that actually matters for future phones: fast, local AI.

Key results from Samsung’s latest tests (via @SemiconductorsX):

• Mobile-BERT (NLP inference): 1199.57 QPS – 2.1x better than Exynos 2500
• Stable Diffusion (image gen): 0.53 QPS – 2.4x improvement

This lines up with Samsung’s earlier internal claim of 113% better generative AI performance on the NPU. It is built on the industry’s first mobile 2nm GAA process, aimed squarely at responsive on-device agentic AI and local generation instead of cloud round-trips.

These are real, public benchmarks (similar to industry-standard MLPerf tests). Samsung’s main point is clear: The Exynos 2600 isn’t just trying to match Snapdragon in regular speed and graphics. Instead, it’s heavily focused on making on-device AI fast, private, and feel instant – all running directly on your phone without needing the internet. Still, these results are encouraging.

Source – Samsung | Via – Yonhap

If you are tired of AI features that only work with a strong internet connection, this is worth watching. Can Samsung finally make Exynos the AI champion in 2026 phones, or will Snapdragon still dominate the flagship market?

The post Samsung Exynos 2600 Benchmarks: 2.1x faster AI & 2.4x stable diffusion on 2nm for Galaxy S26 appeared first on Sammy Fans.