Beijing Yuezhi Dark Noodle Technology Co., Ltd. recently announced that its smart assistant Kimi has ushered in a major upgrade and launched the K1.5 multimodal thinking model. This model surpasses multiple international leading models including GPT-4O and Claude3.5sonnet in terms of multi-mode reasoning and universal reasoning capabilities, and realizes the multi-mode reasoning performance of the company's official version of the official version of O1 for the first time. This breakthrough marks significant progress in the field of artificial intelligence, and the technical innovation and team efforts behind it are worthy of attention. This upgrade not only enhances KIMI's mathematics, code, and visual multi -modal processing capabilities, but also through the strategy optimization method of expanding and improved context, which significantly enhances the deep reasoning capabilities of the model, and provides users with strong support to solve complex problems. Essence
Recently, Beijing Yuezhi Dark Noodle Technology Co., Ltd. announced that its smart assistant Kimi ushered in major technical upgrades and launched a new K1.5 multimodal thinking model. This model has reached the industry -leading level in the multi -mode reasoning and universal reasoning capabilities, marking that Kimi's breakthrough progress in the field of artificial intelligence is marked.
The K1.5 multi -mode thinking model is the third time that Kimi has upgraded its K series enhanced learning model in just three months. After the K0-Math mathematical model released in November last year and the K1 visual thinking model released in December, the K1.5 model performed well in the benchmark test. Under the SHORT-COT mode, K1.5's mathematics, code, visual versatile and general capabilities have greatly surpassed the level of short-term thinking SOTA model GPT-4O and Claude3.5sonnet, with a lead of 550%. In the Long-COT mode, K1.5's mathematics, code, and multi-mode reasoning capabilities have also reached the level of long-term thinking of the official version of the SOTA model Openai O1. Multi -mode reasoning performance.
Behind the upgrade is the unremitting efforts and innovation of the Kimi technical team. For the first time, the team published a detailed model training technology report "Kimi K1.5: With the help of big language models to strengthen the scaling of enhanced learning", it records the road training exploration of model training under the new technology paradigm.
The report pointed out that the key innovation of the K1.5 model includes the expansion of the long up and down context, which improves the training efficiency through some development technology. At the same time, the increase in the length of the contextual length can continue to improve the model performance. In addition, improved strategy optimization methods and simple framework design also provide support for the powerful performance of the model. It is worth noting that the K1.5 model has conducted joint training in text and visual data, which has the ability to combine two modes of reasoning, especially in terms of mathematical ability. There are still challenges.
In order to further enhance the short-chain thinking reasoning ability, the team also proposed a valid Long2short method. Using Long-COT technology to improve the SHORT-COT model, it has achieved significant results in AIME, Math500, and LiveCodebench, which greatly surpasses the existing short short Chain Siwei models such as GPT-4 and Claude Sonnet3.5.
The preview version of the K1.5 multi -mode thinking model will continue to launch the Kimi.com website and the latest version of the Kimi smart assistant APP. If the user finds the model switch button during use, you can experience this new upgrade model. The K1.5 model is good at in -depth reasoning, which can help users solve complex code problems, mathematical problems, and work problems.
Yuezhi Dark Noodle Technology Co., Ltd. stated that in 2025, it will continue to accelerate the upgrade of the K series learning model along the established roadmap, bringing more modes, capabilities in more fields, and stronger universal capabilities to unlock users more for users. possibility.
github report link: https://github.com/moonshotai/kimi-k1.5
The upgrade of the Kimi K1.5 model shows the continuous innovation capabilities of the dark face technology in the field of artificial intelligence. Its leading technical level and future development plan are worth looking forward to. I believe that Kimi will bring users a stronger intelligent experience in the future.