Intel CEO Says Chinese Open-Weight AI Models Are Good Enough

Intel chief executive Lip-Bu Tan said on October 12 that Chinese open-weight AI models are being adopted very quickly because they cost less to run. Speaking at a DBS panel in Singapore, he described models from DeepSeek, Moonshot's Kimi and Zhipu's GLM as good enough for many tasks, according to a Reuters report. The remarks come from a chip executive, which gives them a different angle from the usual debate about whether cheap Chinese models can take share from OpenAI and Anthropic.

What Tan said.

Tan said Chinese frontier models have lower token costs and are seeing fast uptake. He also argued that open source is the best path to efficiency, and that market demand will decide which models suit which use cases. The reporting does not say which benchmarks, if any, Tan was relying on, and it does not describe a specific test comparing the models with U.S. systems.

Chinese open-weight AI models: October pricing.

Third-party pricing compiled by Unwire Pro and summarised by TrendForce puts DeepSeek V4.1 Flash off-peak at $0.15 per million input tokens and $0.60 per million output tokens. Rates double during peak hours. By comparison, GPT-6.1 Sol is listed at $2 input and $10 output, and Claude Opus 5.5 at $4 and $20. Kimi K3 charges $10 per million output tokens, matching GPT-6.1 Sol, while GLM 5.3 charges about 44% of that rate. The same source warns that not every Chinese model is cheap, and that price gaps vary by usage pattern.

Usage on OpenRouter.

A usage figure cited by TrendForce, drawing on the Chinese financial outlet Yicai, says Chinese models processed about 61.2 trillion tokens on OpenRouter in the week of September 7 to 13. That is nearly three times the 21.8 trillion tokens handled by U.S. models. Chinese models have led U.S. models on the platform for 20 consecutive weeks. OpenRouter is only one marketplace, so the figures do not describe the whole market. The Washington Post, also cited by TrendForce, reported that U.S. businesses find open-weight models from Moonshot, DeepSeek and Alibaba cheap to customise, and that DoorDash and Airbnb consider them sufficient for many routine tasks, although U.S. closed models remain more advanced.

The security backdrop.

The adoption story has a security side. Yahoo Finance's account of Tan's remarks notes that on September 8, the NSA, CISA and the FBI accused DeepSeek, Moonshot AI and Z.AI of distilling U.S. frontier models at industrial scale. Distillation means training a smaller model on the outputs of a larger one. Those accusations and the rapid uptake of cheap models now sit side by side in the same policy debate.

Why Intel's view needs context.

Unwire Pro points out that Intel sells chips rather than models. Wider use of low-cost models could raise demand for the hardware that runs them, which gives Tan a commercial reason to praise them. Intel's own results do not isolate any lift from Chinese models either, so the comment is an opinion about market direction rather than a measured effect.

What is still unknown.

The price advantage may not last. Several of the lowest rates come with peak-hour increases, and the Unwire Pro figures are third-party estimates that will change as providers update prices. The Washington Post reports that closed U.S. models are still more advanced, but the reporting does not say how large that gap is on the hardest tasks. Teams that want to test the claim can send the same prompt to several providers and tidy the JSON responses with the free formatter at /en/json-formatter.