Setting the record straight - July 24, 2026

No, there is no new DeepSeek model on July 24, 2026

A rumor is going around: DeepSeek is supposedly releasing a new model on July 24, 2026. That's false. What actually happens that day is the removal of two old API identifiers (deepseek-chat and deepseek-reasoner), redirected to DeepSeek V4-Flash. No new model, no new capability. If you use DeepSeek in the browser or the app, nothing changes for you. If you're a developer, there's just one line to update. Here's the full picture.

Your questions, the real answers

What actually happens on July 24, 2026

The confusion comes from a real technical change that got misread. Here's the clear version.

Is DeepSeek releasing a new model on July 24?

No. There is no announcement of a new model for that date. The latest model is still DeepSeek V4 (V4-Pro and V4-Flash), released on April 24, 2026. The confusion comes from an email and a technical note from DeepSeek about a change on July 24: that change concerns the API, not the model's capabilities. No V5 or R2 version is planned for that date.

I use DeepSeek in the browser or the app: does this affect me?

No, not at all. If you chat with DeepSeek on the website or mobile app, you won't notice any difference. You keep using the same model (V4-Flash by default), for free. The July 24 change only affects developers who call DeepSeek programmatically through the API.

I'm a developer: what do I need to change?

The deepseek-chat and deepseek-reasoner identifiers are removed on July 24, 2026 at 15:59 (UTC). You need to replace them with the new ones: deepseek-v4-flash for standard mode (formerly deepseek-chat) and V4-Flash's reasoning mode (formerly deepseek-reasoner). It's a simple rename of the model identifier in your code. Your API keys and integrations remain valid.

So why is everyone talking about a July 24 release?

Because a technical deadline got relayed on social media as a launch date. It's a classic: "DeepSeek change on July 24" turns into "new DeepSeek on July 24" in a few shares. English-language sources have already corrected it: it's housekeeping in the API, not a roadmap leak.

So when is the next real DeepSeek model?

No official date. A DeepSeek V5 is speculated but has no public roadmap. An R2 (successor to the R1 reasoning model) comes up regularly but has been denied by DeepSeek several times. We track both on dedicated pages, always flagging what's confirmed and what isn't.

Is my data affected by this change?

The July 24 change is purely technical and doesn't alter how your data is handled. That said, the data-privacy question remains relevant at all times with the official app: regulators opened an inquiry as early as January 2025. For sensitive use, prefer a self-hosted or Europe-hosted open-source version.

In one sentence

July 24, 2026 = API cleanup, not a new model

What changes, what doesn't.

What changes (developers only)

  • deepseek-chat is removed, replace it with deepseek-v4-flash
  • deepseek-reasoner is removed, replace it with V4-Flash's reasoning mode
  • Deadline: July 24, 2026 at 15:59 UTC
  • Your API keys and pricing stay unchanged

What doesn't change

  • No new model launches that day
  • The web chat and app work normally, for free
  • V4-Pro and V4-Flash remain the reference models
  • Access stays open, without a VPN

Go further

Understanding the DeepSeek family

The real models, the real changes, and the rumors kept separate from the facts.

DeepSeek V4-Flash

The default model, target of the migration

Learn more

DeepSeek V4 API

Integration guide and up-to-date identifiers

See the guide

DeepSeek V5 rumors

What we know about the next generation

Read the take

DeepSeek R2 rumors

The successor to R1, repeatedly denied

Read the take

Model timeline

From R1 to V4, every version made clear

See the timeline

DeepSeek chat

Test V4 free in the browser

Open the chat

In the meantime

July 24 changes nothing for you: try DeepSeek V4 now

No need to wait for a hypothetical new version. DeepSeek V4 is already here, free in the browser, with a one-million-token context. Test it on a real task and judge for yourself.