the kuaishou kwaipilot team recently officially released two innovative large-scale models, kat-dev-32b and kat-coder, which demonstrate excellent performance in the field of code intelligence. the open-source kat-dev-32b, with 2.3 billion parameters, achieved a 62.4% solution rate in the swe-bench test and ranked fifth among open-source models. it is particularly suitable for lightweight development environments. the closed-source kat-coder achieved a 73.4% solution rate, rivaling the world's top closed-source models and meeting the most stringent development requirements.
both models each have their own advantages, offering developers a wide range of choices. kat-dev-32b is currently available on the hugging face platform, allowing developers to conduct experiments directly. kat-coder provides api services through streamlake, and a technical report will be released at a later date. this differentiated approach meets the needs of the open-source community while also providing high-performance solutions for enterprise users.
against the backdrop of rapid advancements in ai technology, the release of the kat series further promotes the widespread adoption of code intelligence technology. developers can now use more efficient tools to address complex programming scenarios, while kuaishou is able to demonstrate its technological prowess in the ai field.
