Chinese model orders of magnitude cheaper than Western analogs

Post Thumbnail

The Chinese model Kimi K2-Thinking with a trillion parameters cost 4.5 million dollars in the final training stage. According to CNBC, this is orders of magnitude cheaper than Western analogs. The publication cites an anonymous source. And acknowledges that the figure couldn’t be independently confirmed. And Moonshot AI company itself hasn’t officially disclosed the cost.

The model is built on the Mixture of Experts architecture – a trillion parameters in total volume, but only a small part is active during operation. Created for complex reasoning tasks and interaction with tools. Such systems usually require astronomical computing costs, so several million looks almost ridiculous.

History repeats itself. At the end of 2024, Chinese DeepSeek trained the base model V3 for 5.5 million dollars. And the reasoning overlay R1 – for 294 thousand. Kimi K2-Thinking is also built on base K2, so the proportion is similar. However, these estimates don’t include experiments, testing, office rent and salaries.

In most benchmarks K2-Thinking holds at the level of leading Western models like GPT-5 Pro and Grok 4. And now about competitors’ prices. The final training stage of GPT-4, according to SemiAnalysis estimate, cost 63 million dollars. In the AI Index report an even larger sum of 78 million appears. And the full training of Grok 4 was estimated by Epoch AI analysts at astronomical 490 million dollars.

Either this is some magic of optimization, or someone is withholding something. Or someone is greatly overpaying for computations.

AIvengo > Reviews > Chinese model orders of magnitude cheaper than Western analogs
Почитать из последнего
UBTech will send Walker S2 robots to serve on China's border for $37 million
Chinese company UBTech won a contract for $37 million. And will send humanoid robots Walker S2 to serve on China's border with Vietnam. South China Morning Post reports that the robots will interact with tourists and staff, perform logistics operations, inspect cargo and patrol the area. And characteristically — they can independently change their battery.
Anthropic accidentally revealed an internal document about Claude's "soul"
Anthropic accidentally revealed the "soul" of artificial intelligence to a user. And this is not a metaphor. This is a quite specific internal document.
Jensen Huang ordered Nvidia employees to use AI everywhere
Jensen Huang announced total mobilization under the banner of artificial intelligence inside Nvidia. And this is no longer a recommendation. This is a requirement.
AI chatbots generate content that exacerbates eating disorders
A joint study by Stanford University and the Center for Democracy and Technology showed a disturbing picture. Chatbots with artificial intelligence pose a serious risk to people with eating disorders. Scientists warn that neural networks hand out harmful advice about diets. They suggest ways to hide the disorder and generate "inspiring weight loss content" that worsens the problem.
OpenAGI released the Lux model that overtakes Google and OpenAI
Startup OpenAGI released the Lux model for computer control and claims this is a breakthrough. According to benchmarks, the model overtakes analogues from Google, OpenAI and Anthropic by a whole generation. Moreover, it works faster. About 1 second per step instead of 3 seconds for competitors. And 10 times cheaper in cost per processing 1 token.