OpenCode pages list Kimi K4 and DeepSeek V4.1 Pro
Public OpenCode pages name five possible models, including Kimi K4 and DeepSeek V4.1 Pro. Missing specifications and zero token usage limit what the leak proves.
Leak watch. Public listings verified on October 1, 2026. Model release plans remain unconfirmed.
OpenCode has public data pages naming Kimi K4, DeepSeek V4.1 Pro, GLM 5.4, GLM 5.5 Flash and Muse Spark 1.4 Contributor. ByteForward checked all five pages on October 1. They expose model names and provider groupings, while leaving basic product information unknown.
The discovery was already being discussed before today. A September 27 report from Prompt Blueprints attributed the initial sitemap observation to the X account pengsonal. Our check confirms the pages are accessible. It does not independently establish when the sitemap first included them or whether entries were subsequently removed.
What the pages actually show
The Kimi K4 page appears under Moonshot. It reports zero tokens, zero unique users and zero completed sessions in the displayed period ending October 1. Context length, output limits, input capabilities and release timing are all unspecified.
The DeepSeek V4.1 Pro entry has the same missing product fields and reports zero tokens and zero unique users. It also displays 14 completed sessions. That combination deserves caution. Without an explanation of the underlying records, a session counter cannot establish that a working model completed requests.
Under Zhipu, OpenCode lists GLM 5.4 and GLM 5.5 Flash. Both show zero token usage. The Flash page displays one completed session, while the 5.4 page displays none. Neither supplies a release date or usable specification.
The fifth entry is Muse Spark 1.4 Contributor under Meta. It shows two completed sessions alongside zero tokens and zero unique users. The Contributor suffix is part of the visible listing. The page does not explain what access arrangement that name would represent.
A chart can look more certain than its data
There is an important trap in OpenCode’s comparison view. Its DeepSeek V4.1 Pro comparison page displays 50 out of 100 across six capability dimensions, including reasoning and coding. Each value is explicitly described as a neutral placeholder because no data is available.
Those values cannot support a performance ranking. A polished radar chart can make an empty record look like a measured result, so the data label matters more than the shape. The same page leaves pricing, context length and modalities unknown.
The signal and its limits
Taken together, the records make these names worth watching. They show that OpenCode’s public data system contains entries associated with several prominent AI providers. They do not explain who supplied each name, whether the records were entered for testing, or whether the labels match a provider’s intended product name.
That leaves several plausible explanations open, including preparatory integration records and ordinary placeholders. The available pages do not let us choose between them. Their existence also says nothing reliable about launch order. GLM 5.4 and GLM 5.5 Flash appearing together cannot establish which, if either, will arrive first.
For developers, the useful next evidence would be a provider model card, an official API identifier, documented access and pricing, or downloadable weights with a stated license. Until those appear, these listings are a public clue for a watchlist. They are too incomplete to justify a migration plan or a claim that the next frontier model is already available.
Related coverage. Read ByteForward’s report on DeepSeek’s separate Ascend tooling release for the confirmed infrastructure announcement.
Image credit. OpenCode logo from LobeHub Icons, used under the MIT license. Company marks remain trademarks of their owners.
