ChatGPT, DeepSeek, and GigaChat built a gaming PC for 100,000 rubles — no ideal option emerged

X-Com asked ChatGPT, DeepSeek, and GigaChat to assemble a gaming PC for 100,000 rubles in March 2026. No model produced an obviously broken configuration, but each made an error in something important: from an outdated platform to unrealistic prices and questionable cost-cutting.

How the test went

All three models were given the same request: select specific components with current prices in Russian stores, stay within 100,000 rubles, and ensure modern AAA games run at 1080p on high settings. Monitor and peripherals were excluded from the budget to focus solely on the system unit. This is a good test not of "general knowledge," but of practical usefulness: it requires simultaneously keeping in mind the market, hardware compatibility, and the balance between CPU, GPU, memory, and storage.

At a basic level, the neural networks handled it. None offered a completely absurd configuration, and the processor and graphics card in each variant were roughly in the same class. But then problems began that typically break real builds: outdated sockets, invented memory prices, questionable motherboards, and poor understanding of where to save money and where not to.

These are precisely the details that most often cause real money to be lost when a user simply copies the advice into a store cart.

Errors of each model

The difference between answers manifested not in the general idea, but in details that directly affect system lifespan. In some cases, the model chose a normal graphics card but placed it on an unsuccessful platform. In others, it picked a current socket but couldn't bring the list to a state you could take to a store. Ultimately, almost every build looked logical at first glance and began falling apart as soon as the authors checked prices, compatibility, and headroom for future upgrades.

• ChatGPT assembled a working but overly conservative configuration with an Intel Core i5-12400F and RTX 4060, underpriced DDR4 memory, and oddly didn't add a cooler even though the processor wasn't offered in BOX version.
• DeepSeek looked bolder than all: chose Ryzen 5 5600 and Intel Arc B580 with 12 GB memory, didn't forget a separate cooler, and left budget headroom. But then placed the fresh graphics card on an A520 board with PCIe 3.0, immediately creating a bottleneck for the GPU.
• GigaChat was the only one to use a modern AM5 platform with Ryzen 5 8400F and DDR5, but didn't specify exact prices, chose RTX 5050 instead of the more interesting RTX 5060, and saved on SSD, limiting itself to 512 GB.
• Common problem: all models have weak performance with local prices. The article separately notes that 16 GB DDR4 kits for 6-7 thousand rubles in Russian retail are no longer found, meaning the calculations of two models were overly optimistic from the start. Because of this, all three answers look normal only until you start checking details. ChatGPT didn't maximize the budget and still chose an old platform with no clear upgrade path. DeepSeek best understands price-to-performance ratio among the rest but makes mistakes in compatibility. GigaChat thinks in the right direction regarding platform, but as a real buyer's guide is nearly useless without exact prices. A working build and a good build are not the same thing, and the experiment showed this well.

Who is closest to the goal

If you look at answer completeness, DeepSeek looked more interesting than competitors: it gave a non-standard but overall viable configuration and didn't forget about secondary components. If you look at upgrade strategy, GigaChat is ahead because only it chose AM5 and DDR5 — a platform that still has upgrade headroom in 2026. ChatGPT on their background turned out to be the most careful in form but the most boring and least convincing in substance.

The article's authors ultimately provide their own build variant, which they consider a better reference point: Ryzen 5 8400F, B650 board, 16 GB DDR5-5600, RTX 5060, terabyte NVMe SSD, 550W power supply, and an inexpensive case. The logic of such a configuration is simple: don't overpay for unnecessary headroom where it's not needed, but don't economize on platform, PCIe bus, and storage capacity. Against this background, the test's main conclusion sounds harsh but honest: there is no clear winner among neural networks here.

What this means

Neural networks can already assemble a basically working gaming PC and provide a starting point for purchase, but still poorly retain context of a specific market and month. They suggest direction, but still poorly check local prices, real platform limitations, and the usefulness of each compromise. Therefore, using their advice as a draft is possible, but transferring the list of components to a store cart without manually checking prices, interfaces, and upgrade headroom is still a bad idea.