Kimi K2.8 Preview Is Close to K3 — But You Probably Already Have K3

2026-09-12 · 3 min read

kimimoonshotmodel-releasecoding

What's new

On September 11, 2026, Moonshot AI rolled Kimi K2.8 Preview out to every Kimi Code subscriber. The official announcement is restrained — no benchmark table, just four points:

Overall performance is close to K3, with coding and agent capability improved across the board versus K2.7 Code;
Thinking is noticeably more efficient, with low / high / max thinking-effort levels — max by default;
Every subscription tier now gets the full 1M-token context (previously tiered);
The model ID stays kimi-for-coding, so clients and third-party tools need no config change.

That last point is the most practical thing in the release: an unchanged ID means tools like Claude Code and Cline that are already pointed at Kimi Code pick up the new model on their next request, with zero effort from the user. For existing Kimi Code subscribers, it's a free upgrade.

The part the headlines skip: it's subscription-locked

"Close to K3" is easy to misread as "another strong model you can call." It isn't.

K2.8 Preview ships inside the Kimi Code coding subscription, not as a new model ID on the open platform. We checked Moonshot's platform model docs: the current list is kimi-k3, kimi-k2.7-code, kimi-k2.7-code-highspeed and kimi-k2.6. There is no kimi-k2.8-preview and no kimi-for-coding. In other words:

Want to call K2.8 Preview pay-per-token? There's no such option.
Any third-party platform (including us) claiming "K2.8 Preview now available" is either confusing it with K2.7 Code or reselling you a Kimi Code subscription.

That's not a Moonshot quirk; it's the current industry playbook. The newest coding model goes into the vendor's own subscription product first to drive retention, and reaches the API some time later. OpenAI's Codex and Anthropic's Claude Code follow the same rhythm. What you need to recognise as a user is that this tier of model is priced per subscription, not per token.

Where to try Kimi on AIWITH.CHAT

Kimi K3 — the model Moonshot uses as the reference point here, its 2.8-trillion-parameter flagship with native vision and a 1M context — is already on AIWITH.CHAT. No API key, no separate Kimi subscription; it shares one $9.9/month account with GPT-6 Astra, Claude Fable 5.1, Gemini 3.1 Pro and DeepSeek V4.1. The Kimi robot's version dropdown runs from K2.6 through K2.7 Code to K3, so you can compare generations directly.

See plans →

Is it worth switching?

Depends who you are:

You already subscribe to Kimi Code — do nothing; this is a free upgrade and the model ID hasn't changed. One tip: thinking effort defaults to max. If your work is small bug fixes and scripts, dropping to low or high is faster and uses less quota.
You do heavy agentic coding and are happy to pay for a single-tool subscription — Kimi Code is worth considering, but that decision has little to do with K2.8. You'd be buying the subscription and toolchain, not this version number.
You're a general or multi-model user who wants to "try the latest Kimi" — skip this one. K2.8 Preview's official positioning is "close to K3," and K3 is already available on AIWITH.CHAT. Taking out a separate coding subscription to get a preview that aims up at K3 is backwards.

In one line: K2.8 Preview is a free speed-up for existing Kimi Code subscribers, not a model release for everyone. Its reference point, K3, you can use today. The story worth remembering is that the strongest coding models are increasingly shipping only inside subscriptions — not that there's a new model to call.