- Tensor G4: Samsung 4nm, one Cortex-X4 at 3.1GHz, three Cortex-A720 at 2.6GHz, four Cortex-A520 at 1.92GHz, Mali-G715 graphics and a third-generation Google TPU.
- Dimensity 9300: TSMC 4nm, four Cortex-X4 cores and four Cortex-A720 cores with no small efficiency cores, plus Immortalis-G720 MC12 graphics and the APU 790.
- Token claims are not comparable: Google says up to 45 tokens per second for Gemini Nano, a model of about 1.8 to 3.5 billion parameters, while MediaTek says 20 tokens per second for a 7 billion parameter model.
- AnTuTu 11: Pixel 9 Pro about 1.61 million, Pixel 9 Pro XL about 1.67 million, Dimensity 9300 phones 2.32 million to 2.45 million.
- Availability: the Pixel 9 family sells in the United States and Europe. The only Dimensity 9300 phone to reach Europe was the Vivo X100 Pro, in Austria and Czechia at 1,199 euros, and none were sold in the US.
- Google Tensor G4 explained, Android Authority.
- MediaTek Dimensity 9300 product page, MediaTek.
- Generative AI Phone Industry whitepaper, MediaTek.
- Tensor G4 benchmarks under load, Beebom.
- Pixel 9 Tensor G4 heating and throttling tests, Techwiser.
- Tensor G4 benchmarked against rivals, Android Central.
- All the AI features coming to the Pixel 9 series, The Verge.
- Tensor G4 interview on Gemini Nano and tokens, Tom's Guide.
- MediaTek Dimensity 9300 AnTuTu, Geekbench and 3DMark results, Gizmochina.
- Dimensity 9300 chip page, Pixel 9 Pro page and Pixel 9 Pro XL page, NanoReview, checked September 13, 2026.
- Vivo X100 Pro Geekbench 6 result, Geekbench Browser.
- Chipset ranking list with per-chip scores, 91mobiles.
- Pixel 9 launch price and release details, PhoneArena.
- Pixel 9 series European launch pricing and current marketplace prices for the Pixel 9 and Pixel 9 Pro XL, GSMArena.
- Pixel 9 discounted to 499 dollars, Lifehacker.
- Vivo X100 Pro European price and launch, GSMArena, and European availability, PhoneArena.
.png "Google Tensor G4 vs Dimensity 9300 Geekbench Comparison")
Two flagship-class chips arrived within nine months of each other and both were sold on the same word: AI. Google's Tensor G4 powers the Pixel 9 family, launched in August 2024. MediaTek's Dimensity 9300 arrived in November 2023 inside phones from Vivo and Oppo.
On benchmark totals the MediaTek chip is about 1.5 times faster overall, and close to twice as fast in graphics. On AI features that a buyer could switch on, the Pixel wins in most markets, because Google controls the chip, the model, the operating system and the app. That tension is the point of this comparison.
Both chips are now two generations old. The Pixel 9 series is discounted and still widely sold, with software support promised to 2031. Dimensity 9300 phones were mostly China and India models, and only one of them reached Europe. So the practical question in 2026 is not which chip is newer. It is which one still gives you the AI features it promised.
At a Glance
Two bets: Google's TPU against MediaTek's APU
Google built the Tensor G4 around a modest CPU and GPU, then added a third-generation Tensor Processing Unit developed with Google DeepMind. The stated goal was not peak CPU speed. It was running Google's own models on the device, with the software layer designed around them.
MediaTek went the other way. The Dimensity 9300 was the first flagship phone chip with no small efficiency cores at all: four Cortex-X4 cores and four Cortex-A720 cores, all performance class. Its APU 790 was built for generative AI, with support for language models from 1 billion to 33 billion parameters, INT4 quantisation and a Transformer accelerator.
Table 1. Tensor G4 vs Dimensity 9300 on paper
What the AI engines actually deliver
The two chips optimise for different jobs. Google's TPU exists to run one model family well and to feed Pixel features such as Pixel Screenshots, Call Notes and Recorder summaries. MediaTek's APU 790 exists to run many models for whichever phone maker wants them, with no opinion about the app layer above it.
That difference shows up in the software that shipped. Pixel 9 buyers got Gemini Nano with multimodality, meaning the on-device model handles text, images and audio, according to Google and reported by The Verge. Pixel Studio mixes an on-device diffusion model with the cloud Imagen 3 model and needs an internet connection. Add Me uses the chip to stitch two photos into one.
The MediaTek side is not empty, and the split matters. Vivo and Oppo did ship on-device models in China: MediaTek's own generative AI whitepaper records vivo running 1 billion and 7 billion parameter models on device through BlueLM, and Oppo running a 7 billion parameter AndesGPT on the Find X7 series. What those phones did not offer in Western retail was a system-level suite to match Google's.
Table 2. AI features compared
The token number that needs context
Google's headline AI claim for the Tensor G4 was an industry-leading output rate of up to 45 tokens per second. Android Authority's analysis put that figure in context: it comes from running Gemini Nano, a model of roughly 1.8 to 3.5 billion parameters, on the Pixel Screenshots feature.
MediaTek's 20 tokens per second figure is for a 7 billion parameter model, which is at least twice the size. The two numbers therefore do not measure the same workload, and a direct comparison of 45 against 20 would be misleading. What can be said is that Google's TPU is fast for small, tightly controlled models, while the APU 790 is aimed at running larger models through the phone maker's own stack. Google also reserves memory for on-device Gemini, which is one reason the Pixel 9 ships with 12GB of RAM and the Pro models ship with 16GB.
Which phones, and what they score
Benchmark results vary with cooling, software build and sample size, so the table below keeps AnTuTu figures from one aggregator, NanoReview, names the source of every Geekbench entry, and prints the AnTuTu version inside each cell.
Table 3. Scores by phone
Source conflicts to read before the numbers. The Geekbench column for Tensor G4 phones is unstable, because the chip throttles and the samples are small: NanoReview lists 1,268 for the Pixel 9 Pro and 2,001 for the Pixel 9 Pro XL, a gap far wider than any real difference between two phones with the same chip, and Beebom's own Pixel 9 Pro XL measured 1,921 and 3,568 with an AnTuTu 11 total of 1.10 million under load. The Vivo rows have the same problem in miniature, where an X100 sample outscored an X100 Pro sample. Treat each Geekbench cell as one published run, not as a chip specification. AnTuTu 10 and AnTuTu 11 results are not directly comparable either, which is why the version sits inside every cell.
Method for the sub-score comparison: PetaTech24 took AnTuTu 11 sub-scores from NanoReview's Dimensity 9300 chip aggregate and the Pixel 9 Pro device page, both checked on September 13, 2026. The MediaTek figures are CPU 740,213, GPU 789,881, memory 362,038 and UX 555,582, for a total of 2,447,714. The Pixel 9 Pro figures are CPU 593,578, GPU 419,286, memory 209,607 and UX 383,767, for a total of 1,606,238. Dividing gives 1.25 times on CPU, 1.88 times on graphics, 1.73 times on memory, 1.45 times on UX and 1.52 times overall in MediaTek's favour. One caveat applies: the MediaTek numbers are a chip-level aggregate and the Pixel numbers come from a single device page, so read the ratios as a direction rather than a fixed law.
The pattern is consistent with the AI story. The Tensor G4 lags most on graphics and memory, the two areas that gate large on-device models, yet it still delivered more shipped AI features because Google wrote the model and the apps. Raw capacity and usable features are not the same measure.
What is confirmed, what is reported, what nobody knows
Table 4. Status of each major claim
Prices and availability in the US and Europe
Table 5. Devices, availability and prices
US launch prices come from PhoneArena and are consistent with PCMag and Tom's Guide reviews of the same week. The European figures are the widely reported 128GB prices for the Pixel 9 series, carried by GSMArena and other outlets from retail listings ahead of launch. Google does not publish a single Europe-wide price, and local VAT means country prices differ, so treat those euro figures as the launch baseline rather than one fixed store price. The Vivo X100 Pro price for Austria and Czechia comes from GSMArena and PhoneArena.
Street prices have moved a long way since. Lifehacker reported the unlocked Pixel 9 with 128GB at 499 dollars on Amazon in April 2026, against a 799 dollar launch price. For Europe, GSMArena's marketplace listing for the Pixel 9 showed about 431 euros and the listing for the Pixel 9 Pro XL about 523 euros when we checked on September 13, 2026; those are retailer aggregates and vary by country. Google's update commitment for the Pixel 9 runs to 2031, which keeps the AI features current well beyond the hardware generation.
The availability split is the sharpest fact in this comparison. Google sells the Tensor G4 with the full feature set in both the United States and Europe. MediaTek's chip reached Europe in exactly one phone, in two countries, and never had a Western first-party AI suite to match. For buyers in the West, the Dimensity 9300 is largely a benchmark curiosity rather than a purchase option.
How this article was checked
This is a specification and disclosure comparison, not a hardware test. PetaTech24 did not run these phones, these models or these AI features, and no first-party benchmark runs are claimed here. Every number comes from the chip makers, from Google, or from the named third-party tests in the tables, and each table says where its figures come from. Chip-level aggregates and single-device results are labelled as such, and source conflicts are printed instead of averaged away.
FAQ
Is the Tensor G4 better than the Dimensity 9300 for AI?
It depends on what you mean by better. The Dimensity 9300 has more AI capacity on paper, including support for 33 billion parameter models. The Tensor G4, with Gemini Nano, delivered more features that a normal user could actually switch on, including Screenshots, Call Notes and Recorder summaries. Capacity and shipped features are different things.
Which chip is faster overall?
The Dimensity 9300 is faster in every mainstream benchmark, by about 1.5 times overall in the AnTuTu 11 comparison in this article, and by close to twice in graphics. The Tensor G4 trades peak speed for a tighter AI and camera pipeline.
Can I still buy a Pixel 9 in 2026?
Yes. The Pixel 9 remains on sale in the United States and Europe at substantial discounts, and Google's update promise for the model runs to 2031. MediaTek's Dimensity 9300 phones were mostly China and India models, with only the Vivo X100 Pro reaching Europe, so they are hard to buy new in the West.
Do Dimensity 9300 phones run Gemini or other assistants on device?
They run whatever the phone maker ships. The APU 790 supports large on-device language models, and MediaTek listed support for Meta Llama 2, Baichuan and Baidu's Ernie models. In China, vivo and Oppo put their own models on these phones. In Western markets, buyers saw far fewer of those features, because the phones were not sold there and no vendor matched Google's system-level integration.
Bottom line
The Tensor G4 and the Dimensity 9300 are a neat illustration of two AI strategies. MediaTek built more AI capacity into silicon and let partners decide what to do with it. Google built a smaller, controlled stack and shipped features on day one. In benchmarks the MediaTek chip wins by roughly 1.5 times and in graphics by nearly twice, yet in daily use the Pixel 9 remains the phone where the AI features are present, updated and supported until 2031. Neither chip is current in 2026, and buyers should treat both as discounted hardware rather than modern silicon. The lesson carries forward: on-device AI capacity means little until someone ships it.
.png)
.png)
.png)
.png)
.png&description=Tensor G4 vs Dimensity 9300: AI Features Compared in 2026){kind=link}
0 Comments