What happened
On November 2, 2026, a Cursor staff member named deanrie replied in a Cursor community forum thread titled “Pls Add DeepSeek V3.2.” The reply said Cursor now has the hosted model DeepSeek V4.1 Flash. The feature-request thread concerned DeepSeek model support, but Cursor’s reply described V4.1 Flash as the model then available through Cursor’s hosted offering.
Cursor’s forum reply says users can select the model in Cursor’s model picker without supplying their own DeepSeek API key and without configuring Override Base URL. The model is hidden by default; users can enable it in Cursor Settings > Models. The post points readers to Cursor’s Models & Pricing documentation for pricing, but does not state a price in the reply itself. It does not give a rollout date beyond the date of the forum reply, or specify an account tier or geographic availability.
The staff member also describes how the hosted model behaves in Cursor’s Agent: it works like a normal built-in model. That distinction matters because the reply separately discusses users running other DeepSeek options with their own keys. Cursor’s reply says reasoning_content in thinking mode through BYOK does not apply to the hosted model. It does not describe the hosted model’s technical implementation, model parameters, context window, rate limits, or performance.
The reply addresses a routing concern raised in the thread. Cursor says Override Base URL currently affects other models too, and that this is expected. The OpenAI key and override apply to all models except Claude and Gemini. Cursor’s team knows about a request for separate routing, where custom models would go to a user’s endpoint while built-in models remain routed through Cursor, but the reply gives no timeline. For other DeepSeek options such as V4 Pro, the staff member says the request was passed to the team, with no ETA. Users encountering errors with V4.1 Flash are asked to send the Request ID found through the chat menu’s Copy Request ID action.
What the archive adds
New Runtime covered DeepSeek V4.1 Flash becoming available in OpenClaw on October 23, 2026, extending its availability beyond OpenCode. This is a new access point for the same model, not a new model release: Cursor’s hosted integration removes the need for users to bring their own DeepSeek key in this editor.
Trend signal
This is consistent with the direction described in New Runtime’s harness architecture outlives model choice pattern: the editor and its agent experience can become the operational boundary through which a user selects and runs models. The forum reply establishes a built-in selection path, but offers too little evidence to claim a broader change in how Cursor or other coding agents handle model choice.
Why now
Cursor’s reply came in a feature-request discussion asking for DeepSeek support. It presents V4.1 Flash as an option already available in the product, while separately noting that other requested DeepSeek models have no announced timeline. That makes this an incremental response to model-selection demand, not a promised roadmap for the rest of the DeepSeek lineup.
Why it matters
For a Cursor user who wants to try DeepSeek V4.1 Flash, hosted access removes two setup tasks: obtaining and configuring a separate API key, and setting a base-URL override. The model can be enabled in settings and used through the same Agent surface as a built-in model. That lowers the friction of comparing it with the other models available in a coding workflow.
The tradeoff is less control over routing than a self-managed endpoint may provide. Cursor’s reply explicitly says the current override affects multiple models, and Cursor has not announced separate routing for custom and built-in models. Teams that rely on particular endpoints or need to reason about provider-specific behavior should verify their configuration rather than assume that hosted and BYOK use are interchangeable. Pricing is another decision point: Cursor points to its pricing documentation, but the forum reply does not quote rates or limits. The useful operational takeaway is narrower: the hosted option is easier to start, while cost, limits, and routing still need to be checked against the team’s requirements.
Limits and uncertainty
The announcement does not state the hosted model’s price, quotas, rate limits, context window, performance, or availability by plan or region. It does not explain where requests are processed or provide a technical comparison with BYOK. Cursor’s note that the model behaves like a normal built-in model in Agent is a product-level description, not an independent evaluation. Separate routing for custom models remains a known request without an ETA, and the reply gives no schedule for V4 Pro or other DeepSeek options.
Reader takeaway
Cursor users can now try DeepSeek V4.1 Flash through the model picker without bringing their own API key or setting an override. The model must first be enabled in settings. Before adopting it for a team workflow, check Cursor’s current pricing details and verify that the hosted route meets your needs; do not assume BYOK behavior or separate endpoint routing applies.