Kimi’s documentation has turned the K3 launch into a deadline for developers still depending on older Moonshot APIs.
The current Kimi model page says kimi-k2.5 and moonshot-v1 are no longer available to newly registered users. It also says the full platform sunset is scheduled for August 31, 2026, and directs users to switch to newer models.
That makes Kimi K3 more than a headline model. It is now the migration target for teams that built against older Kimi and Moonshot endpoints.
The migration risk is compatibility
Kimi K3 is described in Kimi’s docs as a 2.8 trillion-parameter model with native visual understanding, a 1 million-token context window, and positioning around software engineering, knowledge work, and reasoning.
Those capabilities are useful, but migration risk rarely comes from the headline feature list. It comes from prompt drift, tool-call formatting, response length, latency, rate limits, content filtering, token accounting, SDK assumptions, and whether old workflows depended on quirks that newer models no longer preserve.
Teams should therefore treat the August 31 sunset as a production migration date, not just a documentation notice. A safe plan needs regression prompts, tool-call tests, long-context samples, cost comparisons, and fallbacks for routes that cannot move cleanly.
The old platform is closing in stages
The Kimi docs show a staggered cleanup across older model families. Some prior models already carry discontinued dates, while K2.5 and moonshot-v1 are now blocked for new users and moving toward the full-platform sunset.
That phased pattern is common when labs consolidate after a major generation change. It reduces support burden and nudges developers toward the new architecture. It can also break long-running applications if teams assumed model aliases would remain stable indefinitely.
For Kimi users, the practical question is whether the new K3 path preserves the parts they actually rely on: long context, coding behavior, multimodal inputs, agentic tool use, and regional deployment needs.
