Last verified: August 26, 2026.
Verification warning: DeepSeek’s own documentation is internally inconsistent. Its dated updates page says V4 Flash entered formal release on July 31 and V4 Pro entered formal release on August 13. However, the current pricing and API guides still conflict with that newer release record on prices, Responses API coverage, and V4 Pro reasoning-effort behavior. This page therefore marks those details Unable to verify instead of selecting whichever official page produces the cleaner comparison.
MiniMax vs DeepSeek: the short answer
Choose MiniMax when your project needs a broader creation platform across text, speech, image, and video, or when its specific model behavior wins a matched test. Choose DeepSeek V4 Flash for the lower-latency V4 text route and DeepSeek V4 Pro for the heavier V4 text route—provided your integration test confirms the features you need. Do not choose either vendor from this page on an asserted current DeepSeek price advantage: DeepSeek’s official pricing sources disagree as of August 26, so an exact current cost ranking is not defensible.
The entire V4 family should no longer be called “Preview.” DeepSeek’s official change log records a formal V4 Flash release on July 31 and a formal V4 Pro release across app, web, and API on August 13. A separate deepseek-v4-flash-vision-exp route arrived on August 21 as an experimental visual model; it does not make the base Flash or Pro text routes automatically multimodal.
| Claim | Status on August 26, 2026 | Editorial wording |
|---|---|---|
| V4 Flash release status | Verified | Formally released on July 31 according to DeepSeek’s dated updates page. |
| V4 Pro release status | Verified | Formally released on August 13 across app, web, and API according to the dated updates page. |
| Exact current V4 API prices | Unable to verify | The August update announces a new peak/off-peak schedule effective August 17, while the current pricing page still shows the older table and says peak pricing is forthcoming. |
| Responses API for V4 Pro | Unable to verify | The August 13 update says Flash and Pro support the native Responses API; the current Responses guide still says Flash only and describes Pro as expected in early August. |
| Distinct low reasoning effort on V4 Pro | Unable to verify | The update says low, high, and max for both routes; current API and thinking-mode guides say Pro supports high/max and maps low to high. |
| Visual input on base V4 routes | Unsupported by the reviewed route-specific evidence | Use the separate experimental vision route; do not silently attach its capability to base Flash or Pro. |
Current DeepSeek V4 routes
| Route | Official role | What is safe to say | What still needs a live test |
|---|---|---|---|
deepseek-v4-flash | Faster V4 text route | Formal release is documented; the updates page records native Responses API support. | Exact billed price and any endpoint-specific limits shown to the account. |
deepseek-v4-pro | Heavier V4 text route | Formal release across app, web, and API is documented. | Responses API availability and whether low is a distinct reasoning effort rather than an alias for high. |
deepseek-v4-flash-vision-exp | Experimental visual route | A separately named experimental vision model was announced August 21. | Production suitability, current limits, pricing, and any capability not stated in its model-specific documentation. |
Multimodal work: do not transfer the vision experiment to base V4
DeepSeek’s August 21 update introduced deepseek-v4-flash-vision-exp as a separate experimental visual model. The route name and release note matter: they are not evidence that deepseek-v4-flash or deepseek-v4-pro accept image input, and they do not establish a documented video-understanding or video-generation product. For image testing, call the model-specific experimental route and record its returned model ID, limits, and output. For speech, image generation, or video generation, MiniMax remains the more directly documented platform comparison.
MiniMax’s media breadth also needs one availability qualifier: the current MiniMax model catalog says paid hosted Music and Lyrics API access closed to new users on August 20. Music 3.0 open weights and eligible existing hosted access are different routes.
API compatibility: DeepSeek’s release note and guides conflict
The dated DeepSeek updates page says both V4 Flash and V4 Pro support the native Responses API and low, high, and max reasoning effort as of August 13. Yet the current Responses API guide still says Flash only and describes Pro support as an early-August future addition. The Chat Completions reference and thinking-mode guide also say Pro supports high and max, with low mapped to high.
Editorial decision: formal V4 Pro release is verified by the dated official change log; Pro Responses API support and distinct low effort remain Unable to verify until the route succeeds in a controlled authenticated test or DeepSeek reconciles the guides. Do not turn an announcement-versus-reference conflict into a guaranteed integration claim.
- Pin the exact model route instead of relying on a generic alias.
- Test Chat Completions and Responses separately; success on one endpoint does not prove the other.
- Log the returned model ID, HTTP status, usage fields, and the effective reasoning parameter.
- Keep a fallback when a feature is documented inconsistently.
- Retain the existing FIM caution unless a current model-specific source and test remove that conflict.
Pricing: exact current DeepSeek rates are unable to verify
DeepSeek’s official sources do not currently support one internally consistent price table. The August 13 update announces peak and off-peak pricing effective August 17 and says off-peak rates are half the peak rates. The live pricing page, however, still displays the earlier rates and describes peak pricing as forthcoming. Publishing either table as the unquestioned current rate would hide a first-party conflict.
| Vendor | Safe pricing statement | How to estimate now |
|---|---|---|
| MiniMax | Use the exact model rate in the current official MiniMax pay-as-you-go or token-plan page. | Price the chosen input, cached-input, and output volumes against the dated official table. |
| DeepSeek | Unable to verify exact current public rate because the dated update and current pricing page conflict. | Check the authenticated billing display, run a small metered request, preserve the invoice/usage record, and do not publish the result as a universal rate unless DeepSeek’s public docs align. |
Any older DeepSeek dollar figures may remain only inside a clearly dated historical box such as “pricing observed before the August 17 schedule.” Remove those figures from the current verdict, calculator examples, FAQ, schema, and comparison table. Cost conclusions should be recomputed only after the rate source is reconciled.
Decision guide
| Requirement | Starting point | Gate before production |
|---|---|---|
| Fast text and coding experiments | DeepSeek V4 Flash or the relevant MiniMax text model | Matched quality, latency, endpoint, and metered-cost test. |
| Heavier DeepSeek reasoning route | V4 Pro | Verify the exact endpoint and reasoning-effort behavior your application uses. |
| Image understanding experiment | DeepSeek V4 Flash Vision Exp or an appropriate MiniMax model | Treat DeepSeek’s route as experimental and keep it separate from base V4. |
| Speech, image generation, or video generation | MiniMax | Confirm model-specific availability, inputs, outputs, region, and price. |
| Lowest current API cost | No defensible winner from public docs | Wait for consistent DeepSeek pricing documentation or use a dated authenticated billing test. |
| New hosted music integration | Neither by assumption | MiniMax’s new-user hosted Music/Lyrics access is closed; evaluate open weights or another eligible provider. |
Verdict
DeepSeek V4 is no longer accurately summarized as a preview family: Flash and Pro have dated formal-release entries. The unresolved problem is documentation consistency, not whether Pro was announced. MiniMax is the clearer option for a broad generation stack; DeepSeek remains a credible text and coding candidate, but current pricing, Pro Responses support, and native low effort require verification before they can support a buying claim. A comparison that declares DeepSeek the cheapest today would be more confident than DeepSeek’s own public documentation allows.
Frequently asked questions
Is DeepSeek V4 still in preview?
No as a family-wide description. DeepSeek’s dated updates page records formal V4 Flash release on July 31 and formal V4 Pro release on August 13. The experimental deepseek-v4-flash-vision-exp route is separate and should retain its experimental label.
What is DeepSeek V4 Pro’s current price?
Unable to verify from the current public documentation. DeepSeek’s August update and live pricing page describe incompatible current pricing states. Check authenticated billing and wait for the public sources to align before publishing an exact universal rate.
Does V4 Pro support the Responses API?
Unable to verify without a live account test. The August 13 update says yes, while the current Responses guide still says Flash only and refers to Pro as forthcoming.
Does V4 Pro have low, high, and max reasoning effort?
The official update says all three. The current Chat Completions and thinking-mode references say Pro supports high and max and maps low to high. Treat distinct low effort as Unable to verify until tested or clarified.
Are DeepSeek V4 Flash and Pro multimodal?
Do not infer that from the separate vision release. DeepSeek announced deepseek-v4-flash-vision-exp as its own experimental route. Use model-specific documentation and do not attach its visual capability to the base Flash or Pro IDs.