Related articles

Terminal-Bench 4.0

Coding benchmark where Mistral Large 4 improved from 28% to 41%, narrowly edging GLM-5.3 at 40%; Kimi K3.1 Pro reportedly scores 67%.

• Google announces Gemini 4 Carbon, with a focus on cybersecurity and long-running tasks
• Mistral Unveils 'Le Chonk': A Trillion-Parameter Coding Model That Lags Behind Chinese Rivals