Apple's upcoming A20 Pro chip, expected to power the iPhone 18 Pro, iPhone 18 Pro Max, and the foldable iPhone Duo, has appeared in leaked Geekbench 6 results showing a single-core score of 4,719 and a multi-core score of 12,677. According to Tom's Hardware, this represents roughly a 25% improvement over its predecessor, the A19 Pro. Wccftech's breakdown of the same leak put the gains at 24.5% in single-core and 26.9% in multi-core performance, achieved despite the chip retaining the same core configuration as the A19 Pro (2 performance cores and 4 efficiency cores).
Both outlets note that Apple's own marketing describes the A20 Pro as a 'desktop-class' processor and the 'fastest CPU in a smartphone,' with Apple claiming about a 20% performance improvement over the previous generation — a more conservative figure than what the leaked benchmarks suggest. Tom's Hardware cautions that Geekbench is not necessarily the best indicator of real-world performance and that this is only a single leaked sample.
Wccftech reports that the A20 Pro's single-core score surpasses Apple's M3 chip and beats the M5 Max — the fastest silicon in current MacBook Pros — by about 10% in single-threaded workloads, though the M5 Max still leads by a wide margin in multi-core scores due to having far more cores. Tom's Hardware's comparison table also places the A20 Pro's single-core clock speed at 4.93 GHz, up from the A19 Pro's 4.26 GHz, and shows it outperforming a Qualcomm Snapdragon Elite X2 laptop chip and coming close to an AMD Ryzen 9 9950X3D2 desktop chip in single-core tests. Both outlets attribute these gains to TSMC's 2nm manufacturing process and a redesigned chip package that moves DRAM to the side rather than stacking it on top, allowing better thermal headroom for higher clock speeds.
Separately, Wccftech reports on comments from Max Weinbach on X stating that the A20 Pro's dual 16-core Neural Engine has a higher peak compute throughput than its new 7-core GPU — but only for workloads specifically designed to run on the Neural Engine. Suitable tasks include smaller AI models with around 3 billion parameters, prompt processing during the prefill stage for quantized models, on-device voice recognition, real-time audio enhancement, and vision tasks like camera depth estimation, background blur, text recognition, and image segmentation. Tasks such as unquantized floating-point math, dynamically shaped tensors, and single-token LLM generation are better handled by the CPU or GPU due to Neural Engine dispatch overhead, according to the report.
Wccftech notes that current Geekbench 6 results do not include Metal (GPU) scores yet, and the A20 Pro is the first iPhone chipset to feature a 7-core GPU. Tom's Hardware adds that Apple is expected to open iPhone pre-orders soon, with retail availability set for Friday, September 18, and that more comprehensive third-party benchmarks are likely to follow as review embargoes lift closer to launch.