Sonoma Alpha Sky & Dusk Models: Real-World Performance in Cline

Two AI models, Sonoma Sky Alpha and Sonoma Dusk Alpha, have recently emerged on Vercel AI Gateway and OpenRouter, offering 2 million token context windows and free alpha access, which has sparked interest in the AI community. These models are being evaluated using Cline, a tool that tracks their performance across various coding tasks, revealing that while they are faster, they are less accurate than established models like Claude 4 Sonnet, GPT-5, and Gemini 2.5 Pro, with success rates of 87% and 84% for Dusk and Sky, respectively. Despite their promising specifications and rapid speed, Sonoma models are criticized for issues like hallucinations and tool calling, leading to mixed feedback from users, who are advised to use them for non-critical tasks and rely on more reliable models for production coding. The models represent an intriguing experiment in AI deployment with their free access and mysterious origins, but their current performance indicates they need further development before they can compete with existing models in reliability for real coding work.