Tag: humaneval benchmark

AI

Mistral AI’s Codestral Mamba rollout and benchmark analysis

Mistral AI's 7B Codestral Mamba model achieves 75.0 percent on HumanEval, outperforming DeepSeek v1.5 7B. This Mamba2 architecture offers 0.5 second inference speeds but faces unique security risks from HiSPA triggers identified in 2026 research.