vercel/ai @ai-sdk/deepseek@2.0.64

on GitHub

Patch Changes

• 1b70db6: fix(provider/deepseek): treat the deepseek-flash alias as a V4 model so prior-turn reasoning_content is preserved in multi-turn requests
• 4bd2599: fix(deepseek): preserve reasoning streams across empty tool-call deltas