“Qwen models natively support a 256,000 token context length, which can be extended to 1 million tokens through extrapolation.”