Mistral AI Releases Devstral 2 Coding Models and Mistral Vibe CLI
Mistral
Mistral AI
Mistral AI has launched Devstral 2, a new open-source coding model family available in 123B and 24B sizes, along with a native CLI tool called Mistral Vibe. The models boast state-of-the-art performance on SWE-bench Verified while being cost-effective and deployable on consumer hardware.
Mistral AI released Devstral 2, a next-generation coding model family with two sizes: Devstral 2 (123B parameters) under a modified MIT license, and Devstral Small 2 (24B parameters) under Apache 2.0. Devstral 2 achieves 72.2% on SWE-bench Verified, making it one of the best open-weight models, while Devstral Small 2 scores 68.0% and can run locally on consumer hardware. The models are up to 5x and 28x smaller than DeepSeek V3.2, and 8x and 41x smaller than Kimi K2, offering cost efficiency — up to 7x more cost-efficient than Claude Sonnet at real-world tasks. Mistral also introduced Mistral Vibe CLI, an open-source command-line coding assistant powered by Devstral, which explores, modifies, and executes changes across codebases using natural language. Pricing for Devstral 2 API is $0.40/$2.00 per million tokens (input/output), and $0.10/$0.30 for Devstral Small 2. The models are optimized for data center GPUs (minimum 4 H100-class GPUs for Devstral 2) and consumer GPUs for Devstral Small 2.
- Сокращения
- CLI = Command Line Interface — интерфейс командной строки
- MIT = Massachusetts Institute of Technology — лицензия MIT
- SOTA = State Of The Art — передовой уровень
- API = Application Programming Interface — интерфейс программирования приложений
- GPU = Graphics Processing Unit — графический процессор
Source: Mistral AI —
original