Despite 2M-tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition →context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → and free access, Sonoma Dusk (87%) and Sky (84%) trail Claude 4 Sonnet (96%) on diff edits. Keep a reliable primary model and use these only for non-critical tasks.
Key takeaways · AI-distilled
Cline's figures cover thousands of diff edits from August 26 to September 9, 2025, but the Sonoma models only appeared on September 6, so their share of the sample spans a few days.
Sky is positioned as the stronger reasoning modelA model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.Full definition → and Dusk as the faster one, yet Dusk posted the higher diff-edit success rate; GPT-5 (92%) and Gemini 2.5 Pro (90%) also beat both.
Cline's Discord users reported hallucinations and tool-calling issues with the Sonoma models, even as some praised their speed.
Terms in this piece · Glossary
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
