French AI startup Mistral AI dropped its latest open-source model, Mistral Medium 3.5, on April 29. It's a 128 billion parameter dense architecture, and the company also shipped a remote coding agent tool called Mistral Vibe CLI and a multi-step AI Agent feature for its Le Chat platform.
On paper, the numbers look solid: Mistral Medium 3.5 scored 77.6% on the SWE-Bench Verified coding benchmark, with unified reasoning and coding capabilities. But the community response has been, well, tepid. The reason? Pricing. The API costs $1.5 per million input tokens and $7.5 per million output tokens — significantly higher than what Chinese open-source models charge.
Right now, the open-source model leaderboard is dominated by Chinese entries: Alibaba's Qwen, Zhipu GLM, and Xiaomi's MiMo-V2. Take Qwen 3.6, for example — it has just 27 billion parameters, yet it hits 72.4% on SWE-Bench, and it's completely free and open-source.
Still, some developers argue Mistral's real value isn't raw benchmark scores. It's the "not American, not Chinese" European branding. Because Mistral is based in France, its models comply with GDPR and can be deployed locally within Europe. That's why large institutions like HSBC use Mistral for private deployments. Mistral's European identity is so valued that it landed on TIME's 10 Most Influential AI Companies for 2026, alongside Chinese firms like ByteDance and Zhipu AI. In the end, Mistral's competitive advantage may come down to geopolitics and enterprise compliance, not model performance alone.