DeepSeek API: deepseek-chat and deepseek-reasoner aliases retire July 24 — migration to V4 required

What's new

• Deadline: July 24, 2026 at 15:59 UTC — after this time, deepseek-chat and deepseek-reasoner aliases return errors
• Migrating deepseek-chat: model: 'deepseek-chat' → model: 'deepseek-v4-flash' (fast) or model: 'deepseek-v4-pro' (powerful)
• Critical trap for deepseek-reasoner: The alias maps to deepseek-v4-flash (thinking mode), NOT to V4-Pro. Teams with heavy reasoning workloads must explicitly call deepseek-v4-pro
• What stays the same: Base URL and API key remain unchanged, only the model parameter changes
• Context window: Both V4-Pro and V4-Flash support 1M token context

Why it matters

The 90-day migration window from V4 launch (April 24) is closing. Without migration, all production applications calling legacy aliases will break. The critical trap is deepseek-reasoner → V4-Flash (not Pro), which can degrade performance of reasoning agents.

How to try it

The change is one line per API call. Test both endpoints on your own benchmarks before the July 24 deadline.