French AI company Mistral AI has released the open-source model Mistral Medium3.5, but the market response has been lukewarm

On April 29, French AI startup Mistral AI launched Mistral Medium 3.5—its next-generation open-source model—along with two new tools: Mistral Vibe CLI (a remote coding agent) and Le Chat (a multi-step AI agent feature). The model uses a dense architecture with 1.28 trillion parameters. Official data shows it scored 77.6% on the SWE-Bench Verified programming benchmark and supports unified inference and coding capabilities. However, its API pricing (up to $1.5 per million input tokens and $7.5 per million output tokens) is far higher than Chinese open-source models, drawing widespread community scrutiny. Currently, top open-source model rankings are dominated by Chinese models like Alibaba’s Qwen, Zhifup GLM, and Xiaomi’s MiMo-V2. Notably, Qwen 3.6—with just 270 billion parameters—already hits 72.4% on SWE-Bench and is fully open-source and free. Some developers argue Mistral’s biggest current strength is its “non-U.S., non-China” European identity. Complying with GDPR and European on-premise deployment rules, the model has been privately deployed by major European institutions including HSBC. Market sentiment indicates Mistral’s competitiveness stems more from geopolitics and enterprise compliance than pure model performance.