DeepSeek V4 Pro routes to V4.1 Flash
DeepSeek V4 Pro Now Routes to V4.1 Flash: Migration Guide
- Author
- DeepSeekAgent.io Editorial Team
- Published
- Updated
Since 04:00 UTC on September 14, 2026, requests to deepseek-v4-pro on the official DeepSeek API have followed DeepSeek's announced transition to V4.1 Flash and V4.1 Flash pricing. The temporary arrangement continues until V4.1 Pro launches.
Existing calls remain compatible, while the effective model behind the old ID has changed. From today, production workflows that depended on V4 Pro behavior, quality, or cost should handle this as a model migration.
What the model IDs mean now
| Model ID | Current status | Guidance for new configurations |
|---|---|---|
deepseek-flash | Current V4.1 Flash ID | Recommended |
deepseek-v4-pro | Temporarily routes to V4.1 Flash until V4.1 Pro launches | Compatible with old calls, but cannot pin the former Pro model |
deepseek-v4-flash | Retired and temporarily aliased to V4.1 Flash | Do not use for new configurations |
deepseek-v4-flash-vision-exp | Retired and temporarily aliased to V4.1 Flash | Do not use for new configurations |
Use deepseek-flash in new projects so configuration, logs, and cost attribution stay aligned. Calls using deepseek-v4-pro still succeed, but the name no longer identifies the model being executed.
What changes for existing applications
Rebuild the quality baseline
V4.1 Flash and the former V4 Pro are different models. DeepSeek reports stronger V4.1 Flash results on several Agent benchmarks. Rebuild the production baseline with your own large code changes, long sessions, complex tool use, and strict structured-output workflows.
The route uses Flash pricing
DeepSeek states that deepseek-v4-pro is billed at V4.1 Flash rates during the transition. The total bill still depends on cache hits, the input/output mix, reasoning effort, and Agent turns. Compare quality, runtime, tokens, and the actual charge before and after migration.
The old ID cannot pin the old model
An unchanged ID preserves interface compatibility; the previous V4 Pro weights and behavior are gone from this route. Retain pre-transition outputs, trajectories, parameters, and timestamps wherever historical reproducibility matters.
What DeepSeek Harness users should do
The current DeepSeek Harness model catalog includes deepseek-flash. Select it directly for new Sessions and complete these checks:
- Standardize new tasks on
deepseek-flashin the model selector or Profile. - Record the DSH version, mode, reasoning effort, endpoint, and model ID.
- Run a short task, a cross-file task, and a long tool-driven task as migration smoke tests.
- Compare acceptance criteria, runtime, tokens, tool turns, and manual rework.
- If you use a compatible endpoint, confirm that provider's mapping independently.
What to change now
Make two changes: use deepseek-flash in new configurations, and rebuild monitoring and evaluations around V4.1 Flash. Correct interfaces that still describe deepseek-v4-pro as a stronger Pro tier. For a compatible endpoint, verify the mapping published by that service.
Related reading
- DeepSeek V4.1 Flash overview
- Configure V4.1 Flash in DeepSeek Harness
- Choose DSH Minimal, Standard, or PTC
- V4.1 Flash vs V4 Flash in Harness workloads
FAQ
What model does deepseek-v4-pro return now?
Under DeepSeek's announcement, the route has temporarily served V4.1 Flash since 04:00 UTC on September 14, 2026, and will continue until V4.1 Pro launches.
Is deepseek-v4-pro stronger if I keep the old ID?
The old ID does not preserve the former V4 Pro. It currently routes to V4.1 Flash, so evaluate performance on your own workload.
Which model ID should a new project use?
Use deepseek-flash. It names the current target directly and avoids ambiguity when temporary compatibility routes end.