You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Max mode alone doesn't do it for DeepSeek. Cursor takes the context window from its own model catalog, not from the endpoint, and DeepSeek isn't in that catalog at 1M.
What works: select a catalog model that has a 1M window (e.g. GPT-5.6 Sol). The proxy already rewrites any non-deepseek-* model id to the model: fallback in config.yaml, so DeepSeek answers with the 1M budget Cursor granted. After the first run, a custom deepseek-flash model also gets 1M.
One catch: with a GPT-named model, Cursor sends type: "custom" tools that DeepSeek rejects with a 400. I've opened a PR for that (#53). Until it lands, my fork has the fix, a step-by-step guide and one-click launchers: https://github.com/davidyaaw/deepseek-cursor-proxy-1m
Activity
PauloDaniel1993 commentedon May 26, 2026
enable max mode in cursor
davidyaaw commentedon Sep 24, 2026
Max mode alone doesn't do it for DeepSeek. Cursor takes the context window from its own model catalog, not from the endpoint, and DeepSeek isn't in that catalog at 1M.
What works: select a catalog model that has a 1M window (e.g. GPT-5.6 Sol). The proxy already rewrites any non-
deepseek-*model id to themodel:fallback inconfig.yaml, so DeepSeek answers with the 1M budget Cursor granted. After the first run, a customdeepseek-flashmodel also gets 1M.One catch: with a GPT-named model, Cursor sends
type: "custom"tools that DeepSeek rejects with a 400. I've opened a PR for that (#53). Until it lands, my fork has the fix, a step-by-step guide and one-click launchers: https://github.com/davidyaaw/deepseek-cursor-proxy-1m