Models

← All models

K2-Think

LLM360 · 2025-09-09 · 32.8B parameters

K2-Think is LLM360's 32.8B open-weight reasoning model built on a Qwen2.5-32B base, fine-tuned with supervised chain-of-thought data and reinforcement learning to produce strong extended reasoning capabilities. It was trained and released by LLM360 in collaboration with MBZUAI as a transparent model with its training recipe published. K2-Think demonstrated competitive performance on challenging mathematical benchmarks against models many times its size.

Benchmark scores