Skip to content

deepseek: Add deepseek-v4-pro & deepseek-v4-flash - #54731

Merged
benbrandt merged 2 commits into
zed-industries:mainfrom
cppcoffee:deepseek-v4-pro-flash
Apr 27, 2026
Merged

deepseek: Add deepseek-v4-pro & deepseek-v4-flash#54731
benbrandt merged 2 commits into
zed-industries:mainfrom
cppcoffee:deepseek-v4-pro-flash

Conversation

@cppcoffee

@cppcoffee cppcoffee commented Apr 24, 2026

Copy link
Copy Markdown
Contributor

reference: https://api-docs.deepseek.com/

Closes #54492.

Release Notes:

  • Added deepseek-v4-pro and deepseek-v4-flash models

@cla-bot cla-bot Bot added the cla-signed The user has signed the Contributor License Agreement label Apr 24, 2026
@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch from 32040a9 to ee8529c Compare April 24, 2026 03:14
@lingyaochu

Copy link
Copy Markdown
Collaborator

Maybe we also need to change the context length and the max output tokens.
image

@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch from ee8529c to e9523b7 Compare April 24, 2026 03:25
@cppcoffee

Copy link
Copy Markdown
Contributor Author

Maybe we also need to change the context length and the max output tokens. image

done

@lingyaochu

Copy link
Copy Markdown
Collaborator

#54492 This issue seems related, but I'm not sure if this PR could close that.

@lingyaochu

Copy link
Copy Markdown
Collaborator
image

DeepSeek introduces reasoning effort when in thinking mode, does Zed support this setting? If so, we should also do a change there.

@shenjackyuanjie

shenjackyuanjie commented Apr 24, 2026

Copy link
Copy Markdown
Contributor

I think we should remove the deprcated DeepSeek Chat/Reasoner model name, as they are now route to V4 flash thinking/none thinking, and will be removed later

image

https://api-docs.deepseek.com/

@shenjackyuanjie shenjackyuanjie left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

DeepSeek Chat, DeepSeek Reasoner should be removed

Comment thread docs/src/ai/llm-providers.md Outdated
Comment thread crates/language_models/src/provider/deepseek.rs Outdated
Comment thread crates/deepseek/src/deepseek.rs Outdated
// Their API treats this max against the context window, which means we hit the limit a lot
// Using the default value of None in the API instead
Self::Chat | Self::Reasoner => None,
Self::Chat | Self::Reasoner | Self::V4Flash | Self::V4Pro => Some(384_000),

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

remove chat & reeasoner

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  • The model names deepseek-chat and deepseek-reasoner will be deprecated on 2026/07/24. For compatibility, they correspond to the non-thinking mode and thinking mode of deepseek-v4-flash, respectively.

Would it be better to wait until July to delete it?

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  • The model names deepseek-chat and deepseek-reasoner will be deprecated on 2026/07/24. For compatibility, they correspond to the non-thinking mode and thinking mode of deepseek-v4-flash, respectively.

Would it be better to wait until July to delete it?

I dont think so, as these two name will generate confuse when using

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The old models deleted.

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

From my view point, we should remove the old chat and reasoner model in Zed, because they are only nick names for v4 flash and pro. We couldn't give two models which look different but in fact are the same. Deepseek is the provider, they need to keep the old name for some compatibility, give some time for users (like Zed) to migrate, as the migration is easy for Zed (from I see), we should do it now.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch from 47f9395 to 3a24178 Compare April 24, 2026 04:31
@cppcoffee

Copy link
Copy Markdown
Contributor Author
image DeepSeek introduces reasoning effort when in thinking mode, does Zed support this setting? If so, we should also do a change there.

Effort Control supported.

@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch 2 times, most recently from 341a3cc to 9876a1f Compare April 24, 2026 04:53
@shenjackyuanjie

This comment was marked as outdated.

@lingyaochu

Copy link
Copy Markdown
Collaborator

我想起来个事,既然要移除 chat/reasoner,是不是应该写一个config迁移来着,毕竟这会影响到现有config

I remember something. Since we are removing chat/reasoner, shouldn't we write a config migration? After all, this will affect the existing config

I don't get it, which config are you talking about? Could you please give an example?

@shenjackyuanjie

Copy link
Copy Markdown
Contributor

我想起来个事,既然要移除 chat/reasoner,是不是应该写一个config迁移来着,毕竟这会影响到现有config
I remember something. Since we are removing chat/reasoner, shouldn't we write a config migration? After all, this will affect the existing config

I don't get it, which config are you talking about? Could you please give an example?

image

like in settings file, do we need to translate deepseek-reasoner to deepseek-v4-flash-thinking

@cppcoffee

Copy link
Copy Markdown
Contributor Author

我想起来个事,既然要移除 chat/reasoner,是不是应该写一个config迁移来着,毕竟这会影响到现有config
I remember something. Since we are removing chat/reasoner, shouldn't we write a config migration? After all, this will affect the existing config

I don't get it, which config are you talking about? Could you please give an example?

image like in settings file, do we need to translate deepseek-reasoner to deepseek-v4-flash-thinking

It's fine, users fill in this model name themselves, which is different from the built-in ones.

@shenjackyuanjie

This comment was marked as duplicate.

@SomeoneToIgnore SomeoneToIgnore added the area:ai Related to Agent Panel, Edit Prediction, Copilot, or other AI features label Apr 24, 2026
@cppcoffee cppcoffee changed the title deepseek: add deepseek-v4-pro & deepseek-v4-flash deepseek: Add deepseek-v4-pro & deepseek-v4-flash Apr 24, 2026
@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch 2 times, most recently from abacc75 to 8d5cc11 Compare April 24, 2026 12:26
@cppcoffee

Copy link
Copy Markdown
Contributor Author

Test as follows:

2026-04-24.20.23.57.mov

@shenjackyuanjie

This comment was marked as resolved.

@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch from d3b649e to 219f4e6 Compare April 25, 2026 03:45
@cppcoffee

Copy link
Copy Markdown
Contributor Author

@cppcoffee 还有个事, deepseek现在对有toolcall的是保留全部thinking,无toolcall的 (text thread, 这玩意还有吗) 只保留单次thinking, 这个你做了吗, 我好像没看到相关代码

For turns with tool calls: retain and return all reasoning_content/thinking of that turn.
For turns without tool calls: historical reasoning_content will no longer be included in subsequent context; I have already cleared it before sending it to DeepSeek.

@pandaligx

This comment was marked as spam.

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
@cppcoffee
cppcoffee force-pushed the deepseek-v4-pro-flash branch from 7b65baa to e35d3ae Compare April 25, 2026 07:04
@shenjackyuanjie

This comment was marked as off-topic.

@DJCCRK

This comment was marked as off-topic.

@cppcoffee

This comment was marked as low quality.

@yeskunall

yeskunall commented Apr 27, 2026

Copy link
Copy Markdown
Member

Please do not @- mention team members on your PRs. You guys have polluted this thread with your unnecessary comments. This will be looked at by someone when they get the chance.

@cppcoffee

Copy link
Copy Markdown
Contributor Author

Please do not @- mention team members on your PRs. You guys have polluted this thread with your unnecessary comments. This will be looked at by someone when they get the chance.

Sorry, I mainly really want this feature, after all the Deepseek API is on sale...

@Sherlock-Holo

Copy link
Copy Markdown

will this PR support the new tool call format DSML? I search the word in PR but found nothing...

@cppcoffee

Copy link
Copy Markdown
Contributor Author

will this PR support the new tool call format DSML? I search the word in PR but found nothing...

I did not see any DSML format in the API documentation https://api-docs.deepseek.com/

@Sherlock-Holo

Copy link
Copy Markdown

will this PR support the new tool call format DSML? I search the word in PR but found nothing...

I did not see any DSML format in the API documentation https://api-docs.deepseek.com/

https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/main/encoding/README.md

https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/encoding/README.md

both of them mention the DSML

Co-authored-by: MrSubidubi <dev@bahn.sh>
@benbrandt
benbrandt enabled auto-merge April 27, 2026 13:40
@benbrandt
benbrandt disabled auto-merge April 27, 2026 13:40
@benbrandt
benbrandt enabled auto-merge April 27, 2026 13:40
@benbrandt
benbrandt disabled auto-merge April 27, 2026 13:40
@benbrandt
benbrandt enabled auto-merge April 27, 2026 13:41
@benbrandt
benbrandt added this pull request to the merge queue Apr 27, 2026
Merged via the queue into zed-industries:main with commit 4c27cee Apr 27, 2026
32 checks passed
@benbrandt

Copy link
Copy Markdown
Member

/cherry-pick preview

zed-zippy Bot added a commit that referenced this pull request Apr 27, 2026
…ck to preview) (#55010)

Cherry-pick of #54731 to preview

----
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
tomhoule pushed a commit that referenced this pull request Apr 27, 2026
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
@cppcoffee
cppcoffee deleted the deepseek-v4-pro-flash branch April 28, 2026 07:00
Chillu1 pushed a commit to Chillu1/zed that referenced this pull request Apr 28, 2026
…) (cherry-pick to preview) (zed-industries#55010)

Cherry-pick of zed-industries#54731 to preview

----
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
ebaah46 pushed a commit to ebaah46/zed that referenced this pull request May 6, 2026
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
kathbigra pushed a commit to kathbigra/zed that referenced this pull request May 10, 2026
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
jonx pushed a commit to jonx/zed-aros that referenced this pull request Jul 17, 2026
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
jolutz pushed a commit to jolutz/zed that referenced this pull request Aug 8, 2026
reference: https://api-docs.deepseek.com/

Release Notes:

- Added deepseek-v4-pro and deepseek-v4-flash models

---------

Signed-off-by: Xiaobo Liu <cppcoffee@gmail.com>
Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>
Co-authored-by: MrSubidubi <dev@bahn.sh>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area:ai Related to Agent Panel, Edit Prediction, Copilot, or other AI features cla-signed The user has signed the Contributor License Agreement

Projects

None yet

Development

Successfully merging this pull request may close these issues.

DeepSeek model name & thinking option updated

10 participants