Chinese AI Kimi K3 Closes Gap with US Models
- tech360.tv
- 1 day ago
- 3 min read
Chinese start-up Moonshot AI's Kimi K3 has significantly reduced the artificial intelligence capabilities gap between China and the US, a development prompting concern in Silicon Valley. Beijing based Moonshot AI, a unicorn company, has affirmed its commitment to an open source strategy for AI. The latest model's release marks a notable shift in the competitive landscape of frontier AI.
Moonshot AI introduced the Kimi K3, a substantial large language model boasting 2.8 trillion parameters, recently. The organisation states this model includes native vision abilities and was developed for extended coding tasks, knowledge based work, and intricate reasoning processes. It represents a considerable offering from the Chinese firm.
The Kimi K3 model demonstrated superior performance against established closed source US models such as Anthropic's Claude Opus 4.8 and OpenAI's GPT five point five. It only lagged behind the more recently launched Claude Fable 5 and GPT five point six Sol, according to Artificial Analysis. So, this positions K3 among the leading global AI models.
This performance challenges the long held belief that Chinese models typically lag six to eight months behind their American counterparts. Ryan Fedasiuk, a fellow at the American Enterprise Institute, detailed this in a recent report titled "China has caught up in frontier AI." He stated that this time difference has now shrunk to merely weeks.
Furthermore, Kimi K3 offers a cost advantage. The model costs an average of US$0.95 per task, making it considerably more economical than its US counterparts. For reference, Fable 5 costs US$2.75, and GPT five point six Sol costs US$1.06 per task, as noted by Artificial Analysis.
The advanced abilities of Kimi K3 have not gone unnoticed among US tech executives. Cline, an open source AI coding agent platform, described the launch as a "game changing milestone for open weights" on social media. And Dean Ball, head of strategic futures at OpenAI, recognised K3 as a "very good model" whose performance, he stated, could not be attributed to distillation.
US research organisations, including Anthropic and OpenAI, have frequently suggested that Chinese AI advancements stem from distillation. This industry practice involves training smaller models using outputs generated by larger, more advanced systems. This explanation is now being reconsidered by some experts.
Kyle Chan, a fellow at the US think tank the Brookings Institution, countered this prevailing notion on social media. He stated that Kimi K3 "puts to rest the idea that Chinese AI models are good just because they distil on US models," noting its superior performance over models theoretically available for such a process, like Opus 4.8 and GPT five point five.
Chan further asserted that the model "proves that Chinese AI capabilities are real, not just derived from distillation on American models." But, the rapid progress has heightened US anxieties regarding China's advancements in artificial intelligence. This sentiment was echoed by David Sacks, a former AI adviser under President Donald Trump's administration.
Sacks, who also serves as co chair for Trump's Council of Advisers on Science and Technology, expressed his concern on social media. He criticised current US policies, stating, "America is tying itself in knots: politicians and bureaucrats are banning new data centres, piling on state regulations and pushing for new federal agencies to pre approve frontier models."
OpenAI's Ball also raised concerns about the wider implications of open weight models. He warned that if these models were to dominate the global ecosystem, it could lead to what he termed "full AI communism," a future he described as a "dystopian hellscape." Such commentary highlights the differing views within the Big Tech sector.
Despite Kimi K3's notable capabilities, users have reported issues with its operational speed. Jun Song, founder of the Supergemma Foundation, which funds open source projects via cryptocurrency, commented on social media that K3 was "currently so slow and frustrating to use" despite its strengths. And Moonshot AI acknowledged the performance issues. The organisation issued a statement indicating that substantial demand had strained its computing infrastructure, necessitating a temporary halt to new subscriptions. According to the South China Morning Post, Moonshot AI apologised to potential users and announced plans to increase its computing capacity, with subscriptions set to reopen gradually.
Moonshot AI's Kimi K3 model has reduced the AI capability gap between China and the US to weeks.
The 2.8 trillion parameter model outperformed several premier US closed source models and is cheaper per task.
Industry experts dispute the theory that Chinese AI advancements are solely due to distillation from US models.
Concerns have been raised in the US regarding the implications of China's rapid AI progress and the risks of open weight models.
Moonshot AI has temporarily suspended new subscriptions for Kimi K3 due to overwhelming demand and strained computing resources.
Source: SCMP
