[Bug] DeepSeek-V4.1-Flash display name misses the decimal point — still present in 0.2.0-rc.2

[zyiyi-dev]

Follow-up to #6114 — still reproducible after the recent refactor.

Environment

• @deepseek-ai/dsh 0.2.0-rc.2 (desktop app, macOS arm64, nightly build 2026-09-29T10:18Z)
• The selector row and its hover tooltip both show DeepSeek-V41-Flash

Where it lives now
The catalog was split out of index.ts since the original report. The string is now at
packages/llm/llm-deepseek/src/models.ts:9 — name: 'DeepSeek-V41-Flash' — completely
unchanged. Line 17 of the same array uses the correct style, name: 'DeepSeek-V4-Pro',
so the two entries are inconsistent with each other.

Expected
DeepSeek-V4.1-Flash, matching the official release note:
https://api-docs.deepseek.com/news/news260910/ ("DeepSeek-V4.1-Flash",
"Set your model to deepseek-flash").

Scope
Display only. The model id deepseek-flash is correct — requests, billing and capabilities
are unaffected. The selector renders model.name verbatim, so this single string is the
entire source of the label.

Possible cause / fix
packages/llm/llm-deepseek/src/image-tokens.ts:4 refers to this generation as
"its current v41 configuration", so the internal shorthand seems to have leaked into the
user-facing name. One-character fix: V41 → V4.1.

[slyyr]

Since the incorrect name appears both in the selector and tooltip, I’d look for the shared model-catalog/display-name definition rather than fixing each UI component separately.

If the catalog was recently moved out of index.ts, there may be a normalization or formatting function that is stripping punctuation from the canonical model name.

I’d search for the exact string:

DeepSeek-V41-Flash

and then trace backwards to determine whether it is stored incorrectly or generated incorrectly.

Ideally the model metadata should keep a canonical display name:

DeepSeek-V4.1-Flash

and the UI should render that value directly rather than reconstructing the name from an internal model ID.

I’d also add a small regression test around the model catalog asserting that the display name remains exactly DeepSeek-V4.1-Flash.

That would prevent another formatter/refactor from silently changing the user-facing model name again.

If this clears up the issue, please consider marking the answer as accepted so others who run into the same problem can find the solution more easily.

[slyyr]

Since the incorrect name appears both in the selector and tooltip, I’d look for the shared model-catalog/display-name definition rather than fixing each UI component separately.

If the catalog was recently moved out of index.ts, there may be a normalization or formatting function that is stripping punctuation from the canonical model name.

I’d search for the exact string:

DeepSeek-V41-Flash

and then trace backwards to determine whether it is stored incorrectly or generated incorrectly.

Ideally the model metadata should keep a canonical display name:

DeepSeek-V4.1-Flash

and the UI should render that value directly rather than reconstructing the name from an internal model ID.

I’d also add a small regression test around the model catalog asserting that the display name remains exactly DeepSeek-V4.1-Flash.

That would prevent another formatter/refactor from silently changing the user-facing model name again.

If this clears up the issue, please consider marking the answer as accepted so others who run into the same problem can find the solution more easily.