Kimi K2.8 Preview Brings One-Million-Token Context to Code
Moonshot has upgraded Kimi Code with a K2.8 preview that keeps the same model ID, adds adjustable reasoning, and opens a million-token context window.
Moonshot AI has rolled out Kimi K2.8 Preview across Kimi Code, upgrading the existing kimi-for-coding route without requiring users or third-party clients to change configuration. The preview is available across membership tiers and supports context windows of up to one million tokens.
A larger default coding workspace
Kimi says K2.8 Preview approaches the performance of its K3 flagship while using reasoning more efficiently than K2.7 Code. It adds three selectable thinking levels—low, high and max—with max set as the default. The model also accepts image and video inputs, extending Kimi Code beyond text-only repository work.
The unchanged model ID is strategically important. Rather than asking developers to test and migrate to another endpoint, Moonshot has made the newer system the default path behind an established name. Existing workflows, editor integrations and command-line setups can therefore receive the upgrade without a visible migration step.
The million-token window is the headline capability. In principle, it allows a coding agent to hold more source files, specifications, logs and prior decisions in one working context. That could reduce the need to repeatedly summarize large repositories, particularly during debugging or multi-file refactoring. It does not, however, prove that the model can reliably use every token: retrieval quality, attention stability, latency and cost remain practical constraints.
Why it matters
K2.8 gives Moonshot a fresh way to compete in coding agents during an otherwise quiet Asian news cycle. The release also narrows the product gap between a company’s flagship model and its more accessible coding route. The important evidence still missing is independent evaluation on sustained repository tasks, where a million-token limit matters only if the agent can select the right evidence and act consistently over many steps.