K2-V2: A 360-Open, Reasoning-Enhanced LLM

TL;DR

K2-V2 is a fully open, reasoning-enhanced large language model that outperforms many existing models in size and capability, with a focus on reasoning, domain knowledge, and tool use, and is openly released for community use.

Contribution

The paper introduces K2-V2, a new open-source LLM with superior reasoning and performance, and provides full training data and artifacts to support community development.

Findings

Outperforms Qwen2.5-72B and approaches Qwen3-235B performance.

Explicitly infused reasoning, domain knowledge, and tool use during training.

Open release of training data and model weights to facilitate further research.

Abstract

We introduce K2-V2, a 360-open LLM built from scratch as a superior base for reasoning adaptation, in addition to functions such as conversation and knowledge retrieval from general LLMs. It stands as the strongest fully open model, rivals open-weight leaders in its size class, outperforms Qwen2.5-72B and approaches the performance of Qwen3-235B. We actively infuse domain knowledge, reasoning, long-context, and tool use throughout the training process. This explicitly prepares the model for complex reasoning tasks. We demonstrate this potential using simple supervised fine-tuning, establishing a strong baseline that indicates significant headroom for advanced alignment. By releasing the full training history and data composition, we maximize the effectiveness of continuous training, a key open source production scenario. We release the model weights and signature LLM360 artifacts, such as complete training data, to empower the community with a capable, reasoning-centric foundation.