Google's Tensor G6 is a bigger shake-up than a normal yearly refresh. It swaps in Arm's new C1-series CPU cores, adds a fifth-generation TPU, and moves to a MediaTek modem in place of Samsung's. To see what that means in practice, this assessment puts the Tensor G6 (Pixel 11 Pro XL) head to head with last year's Tensor G5 (Pixel 10 Pro) across CPU, GPU, sustained load and on-device AI.
What Changed on Paper
Both chips are built on TSMC 3nm, with the Tensor G6 using the enhanced N3P version of the process. The real change is the CPU. The Tensor G5 used a 1+5+2 layout of Cortex-X4, Cortex-A725 and Cortex-A520 cores. The Tensor G6 drops to seven cores in a 1+4+2 layout, built from Arm's newer C1-Ultra prime core at 4.11GHz, four C1-Pro cores at 3.38GHz and two more C1-Pro cores at 2.65GHz.
The GPU stays in the PowerVR family but moves from the D-series DXT-48-1536 at 1.1GHz to the C-series CXTP-48-1536 at 1.29GHz. The TPU goes from fourth to fifth generation, with Google claiming 50% more compute and higher memory bandwidth. Security moves from the Titan M2 to the Titan M3, and the modem changes from Samsung's Exynos 5400 to the MediaTek M90.
Overall Performance: AnTuTu
The Tensor G6 scored 1,687,489 in AnTuTu against 1,291,252 for the Tensor G5, a lead of roughly 31%. The category breakdown is where the story gets interesting. CPU is up about 62%, memory is up about 40% and UX is up about 71%, but the GPU score fell by about 34%.
CPU Performance: Geekbench 6
Geekbench 6 shows the new prime core doing its job. Single-core climbs from 2,285 to 2,691, about 18% faster, while multi-core rises from 6,191 to 7,203, about 16% faster. The multi-core gain is smaller than the single-core gain because the Tensor G6 has one fewer core to throw at parallel work.
Sustained Performance and Throttling
A 15-minute CPU throttling test, which keeps every core under heavy load, is where the Tensor G6 gives some ground back. The Tensor G5 peaked at 408,383 GIPS while the Tensor G6 peaked at 343,713 GIPS, about 16% lower. By the end of the run the Tensor G6 had dropped to 47% of its peak, while the Tensor G5 held on to 61%.
The Pixel 11 Pro XL did not get abnormally hot during the run, and both phones use a vapour chamber, so heat alone does not explain the gap. The more likely explanation is that Google is capping clock speeds under sustained load to keep the phone cool and efficient. In everyday use that should feel fine, but it does mean the newer chip is not the better choice for long, all core workloads.
GPU: The Weak Spot
This is the one area where the Tensor G6 does not move forward. Despite the higher clock speed and the newer C-series design, the AnTuTu GPU score drops from 382,578 to 251,126. Anyone buying a Pixel mainly for demanding 3D games will not see a graphics upgrade over the Pixel 10 Pro, and the chip still sits well behind the best rival silicon in gaming.
On-Device AI
Tensor chips are built around on-device AI, so this is the most important test for the platform. Running the Gemma-4-E2B-it language model on the CPU, the Tensor G6 reached a prefill speed of 271.48 tokens per second, about 64% faster than the Tensor G5. Decode speed, which controls how quickly the answer is actually written out, more than doubled from 10.42 to 24.67 tokens per second, a gain of about 137%. Time to first token also improved from 1.65 seconds to 0.98 seconds.
These numbers do not use the TPU at all, since third-party apps cannot access it yet. That means they reflect the CPU upgrade alone. Google claims a 3.5x improvement in both speed and energy efficiency for on-device AI tasks, which is something CPU-only testing cannot confirm.
Where the Tensor G6 Wins and Loses
Putting every result on one scale makes the trade-off clear. The Tensor G6 wins big in AI, UX and CPU workloads, and loses in GPU and in the sustained, all-core scenarios.
Scorecard
Verdict
The Tensor G6 is a clear step forward where it counts for most Pixel owners. The CPU is faster, the system feels quicker, and on-device AI responses are generated more than twice as fast even before the TPU is involved. Google has clearly prioritised efficiency and thermals over peak numbers, which is why sustained load and GPU results trail the Tensor G5.
If you are upgrading from a Pixel 10, the Pixel 11 will feel faster and smarter, but gamers should not expect a graphics upgrade. For everyone else, this is the most meaningful Tensor generation in a while.
Benchmark figures in this article come from side-by-side testing of the Pixel 11 Pro XL and Pixel 10 Pro published by Beebom. Results vary between devices and software versions.
