Kimi K4 in OpenCode Has Not Yet Been Confirmed by Moonshot AI

What is actually confirmed about Kimi K4

I would not call Kimi K4 a new public release yet. As of September 26, 2026, the available official Moonshot AI materials contain no standalone announcement or model card for it. An OpenCode data page carrying that name does not, by itself, confirm a new API, the model's existence, or its claimed specifications—especially when the primary source is silent.

The Kimi API platform documentation lists interfaces compatible with OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages. Kimi Code documentation also describes dedicated endpoints for coding tasks and different base URLs for China and other regions. These are real integration surfaces, but they do not turn a K4 mention into an official release.

The newest extensively documented model in the Moonshot AI materials found here is Kimi K3. Its organization overview cites 2.8 trillion parameters, Kimi Delta Attention, Attention Residuals, native image support, and a one-million-token context window. OpenCode listings for K3 also show pricing and quotas, while setup guides reference the K2.6 and K2.5 identifiers.

Connecting Kimi to OpenCode is nevertheless documented quite specifically. One path uses the opencode auth login command and the Kimi For Coding provider; another requires a Moonshot AI key and provider selection within OpenCode. Provider availability, however, is not confirmation of every model name that appears in a catalog or page URL.

Why the model name matters here

The practical conclusion is straightforward: it is too early to build technical expectations around Kimi K4. Without a model card, there is no reliable way to establish its context window, modalities, limits, pricing, quotas, or even its exact API identifier. For an integration, those are not minor details; they are the contract governing request routing, cost estimates, and error handling.

Kimi's compatible interfaces do reduce friction when connecting existing SDKs and agent tools. Yet transport compatibility says nothing about a particular model's quality, stability, or behavior on long-running tasks. I would first verify the identifier against Moonshot AI's official model list, then verify its capabilities against a dedicated model card.

For now, Kimi K4 looks less like a confirmed new milestone and more like a name that has outrun the documentation. An official announcement may appear later, but the gap between the OpenCode page and primary-source materials remains the central fact of this story.

We previously examined Claude Opus 4.6 through its intelligence, pricing, configuration, and context-cost trade-offs. That comparison helps place Kimi K4’s OpenCode release in the wider landscape of models available to developers.