\n\ud83d\ude80 Hello, Kimi K2 Thinking!
The Open-Source Thinking Agent Model is here.<\/p>\n\ud83d\udd39 SOTA on HLE (44.9%) and BrowseComp (60.2%)
\ud83d\udd39 Executes up to 200 \u2013 300 sequential tool calls without human interference
\ud83d\udd39 Excels in reasoning, agentic search, and coding
\ud83d\udd39 256K context window<\/p>\nBuilt\u2026 pic.twitter.com\/lZCNBIgbV2<\/a><\/p>\n
\u2014 Kimi.ai (@Kimi_Moonshot) November 6, 2025<\/a><\/p><\/blockquote>\n
Performance metrics challenge US models<\/p>\n
According to the company\u2019s GitHub blog\u00a0post<\/a>, Kimi K2 Thinking scored 44.9% on Humanity\u2019s Last Exam, a large language model benchmark consisting of 2,500 questions across a broad range of subjects, exceeding GPT-5\u2019s 41.7%.<\/p>\n
The model also achieved 60.2% on the BrowseComp benchmark, which evaluates web browsing proficiency and information-seeking persistence of large language model agents, and scored 56.3% to lead in the Seal-0 benchmark designed to challenge search-augmented models on real-world research queries.<\/p>\n
VentureBeat\u00a0reported<\/a>\u00a0that the fully open-weight release meeting or exceeding GPT-5\u2019s scores marks a turning point where the gap between closed frontier systems and publicly available models has effectively collapsed for high-end reasoning and coding.<\/p>\n
\nKimi K2 Thinking is the new leading open weights model: it demonstrates particular strength in agentic contexts but is very verbose, generating the most tokens of any model in completing our Intelligence Index evals@Kimi_Moonshot<\/a>‘s Kimi K2 Thinking achieves a 67 in the\u2026 pic.twitter.com\/m6SvpW7iif<\/a><\/p>\n
\u2014 Artificial Analysis (@ArtificialAnlys) November 7, 2025<\/a><\/p><\/blockquote>\n
Cost efficiency raises\u00a0questions<\/p>\n
The popularity of the model grew after CNBC reported its training cost was merely US$4.6 million, though Moonshot AI did not comment on the cost.\u00a0According to calculations by the\u00a0South China Morning Post<\/a>,\u00a0the cost of\u00a0Kimi K2 Thinking\u2019s application programming interface was\u00a0six\u00a0to 10 times cheaper than\u00a0that\u00a0of OpenAI and Anthropic\u2019s models.<\/p>\n
The model uses a Mixture-of-Experts architecture with\u00a0one\u00a0trillion total parameters, of which 32 billion are activated per inference, and was trained\u00a0using\u00a0INT4 quantisation to achieve roughly\u00a0two times\u00a0generation speed\u00a0improvement\u00a0while maintaining state-of-the-art performance.<\/p>\n
Thomas Wolf, co-founder of Hugging Face,\u00a0commented<\/a>\u00a0on X that Kimi K2 Thinking was another case of an open-source model passing a closed-source model, asking, \u201cIs this another DeepSeek moment? Should we expect [one] every couple of months now?\u201d<\/p>\n
Technical capabilities and limitations<\/p>\n
Moonshot AI researchers\u00a0said<\/a>\u00a0Kimi K2 Thinking set \u201cnew records across benchmarks that assess reasoning, coding and agent capabilities\u201d. The model can execute up to 200-300 sequential tool calls without human interference, reasoning coherently across hundreds of steps to solve complex problems.<\/p>\n
Independent testing by consultancy Artificial Analysis placed Kimi K2 on top of its Tau-2 Bench Telecom agentic benchmark with 93% accuracy, which was\u00a0described<\/a>\u00a0as the highest score it has independently measured.<\/p>\n
However, Nathan Lambert, a researcher at the Allen Institute for AI, suggested there\u2019s still a time lag of approximately four to six months in raw performance between the best closed and open models, though he\u00a0acknowledged<\/a>\u00a0that Chinese labs are closing in and performing very strongly on key benchmarks.<\/p>\n
Market implications and competitive pressure<\/p>\n
Zhang Ruiwang, a Beijing-based information technology system architect, said the trend was for Chinese companies to keep costs down, explaining, \u201cThe overall performance of Chinese models still lags behind top US models, so they have to compete in the realms of cost-effectiveness to have a way out\u201d.<\/p>\n
Zhang Yi, chief analyst at consultancy iiMedia, said the training costs of Chinese AI models were seeing a \u201ccliff-like drop\u201d driven by innovation in model architecture and training technique, and input of quality training data, marking a shift away from the heaping of computing resources in the early days.<\/p>\n
The model was released under a Modified MIT License that grants full commercial and derivative rights, with one restriction: deployers serving over 100 million monthly active users or\u00a0generating<\/a>\u00a0over US$20 million per month in revenue must prominently display \u201cKimi K2\u201d on the product\u2019s user interface.<\/p>\n
Industry response and future outlook<\/p>\n
Deedy Das, a partner at early-stage venture capital firm Menlo Ventures, wrote in a post on X that \u201cToday is a turning point in AI. A Chinese open-source model is #1. Seminal moment in AI\u201d.<\/p>\n
\n\ud83d\udea8 Today is a turning point in AI. A Chinese open source model is #1.<\/p>\n
Kimi K2 Thinking scored 51% in Humanity’s Last Exam, higher than GPT-5 and every other model. $0.6\/M in, $2.5\/M output.<\/p>\n
The best at writing, and does 15tps on two Mac M3 Ultras!<\/p>\n
Seminal moment in AI.<\/p>\n
Try it\u2026 pic.twitter.com\/fmxlxpCGbE<\/a><\/p>\n
\u2014 Deedy (@deedydas) November 7, 2025<\/a><\/p><\/blockquote>\n
Nathan Lambert wrote in a Substack article that the success of Chinese open-source AI developers, including Moonshot AI and DeepSeek, showed how they \u201cmade the closed labs sweat,\u201d adding \u201cThere\u2019s serious pricing pressure and expectations that [the US developers] need to manage\u201d.<\/p>\n
The release positions Moonshot AI alongside other Chinese AI companies like DeepSeek, Qwen, and Baichuan that are increasingly challenging the narrative of American AI supremacy through cost-efficient innovation and open-source development strategies.\u00a0<\/p>\n
Whether this represents a sustainable competitive advantage or a temporary convergence in capabilities remains to be seen as both US and Chinese companies continue advancing their models.<\/p>\n
the public nature of the statements, and the market\u2019s reaction, suggest substantive discussions may soon be underway.<\/p>\n
The AI chip landscape is entering a period of flux. Organisations should maintain flexibility in their infrastructure strategy and monitor how partnerships like Tesla-Intel might reshape the competitive dynamics of AI hardware manufacturing.<\/p>\n
The decisions made today about chip manufacturing partnerships could determine which organisations have access to cost-effective, high-performance AI infrastructure in the coming years.<\/p>\n
Photo by Moonshot AI<\/a>)<\/p>\n
See also:<\/strong> DeepSeek disruption: Chinese AI innovation narrows global technology divide<\/a><\/p>\n
<\/a><\/p>\n
Want to learn more about AI and big data from industry leaders?<\/strong> Check out AI & Big Data Expo<\/a> taking place in Amsterdam, California, and London. This comprehensive event is part of TechEx<\/a> and co-located with other leading technology events. Click here<\/a> for more information.<\/p>\n
AI News is powered by TechForge Media<\/a>. Explore other upcoming enterprise technology events and webinars here<\/a>.<\/p>\n