Kimi K2 API Retirement: What Developers Must Change Before December Ends

The Kimi K2 API retirement is now official: Moonshot AI stopped offering K2 models to new accounts on October 31, 2026, and plans to shut the whole K2 series down by the end of December. If your app calls a K2 model today, you have about two months to move.

The good news is that the switch is mostly a model-name change. The catch is cost and one default setting that can quietly make your requests slower and pricier.

Here are the facts at a glance:

• New accounts lost K2 access on October 31, 2026; existing users are not affected yet.
• Moonshot plans to retire all K2 models by the end of December 2026.
• Moonshot points developers to a migration guide for Kimi K3.1 Pro.
• K3 models offer a 1,048,576-token context window, up from 262,144.

Those dates matter because existing keys keep working only until the shutdown, so waiting costs you nothing now but leaves little room to test later.

What the Kimi K2 API retirement means for your app

Moonshot’s own API pricing page carries the notice. It says K2 series models are no longer available to newly registered users from October 31, and that the company plans to retire all of them by the end of December 2026.

If you built your product on a K2 model and your account already exists, nothing breaks today. But the clock is running, and new teammates or new projects on fresh accounts cannot use K2 at all.

Per KissAPI’s migration guide, published October 24, the affected models are kimi-k2.7-code, kimi-k2.7-code-highspeed and kimi-k2.6. The same guide says kimi-k2.5 and moonshot-v1 were already discontinued on August 31. That is a third-party summary, so check your own dashboard for the exact models on your account.

Move to Kimi K3.1 Pro: what changes

Moonshot does not publish an official one-to-one model map, but its notice steers developers to Kimi K3.1 Pro. Per the KissAPI guide, the main K2.7 Code workload only needs the model string swapped, and the OpenAI SDK works with a base URL change.

We covered the model itself in our Kimi K3.1 Pro price and benchmarks post. This one is about the practical switch.

The table below compares the models, using prices from the KissAPI guide and the context sizes from Moonshot.

Going from K2.7 Code to K3.1 Pro raises input cost about 2.1x and output cost about 2.5x, by the guide’s math. The bigger context window can offset that if you currently split long documents into chunks.

The reasoning effort setting that can raise your bill

This is the part most migration notes skip. According to the KissAPI guide, K3.1 Pro always thinks: you cannot turn thinking off, and reasoning_effort accepts low, high or max, with max as the default.

So if you only change the model name, every request runs at max effort. That means more output tokens and slower replies than you were used to on K2.

A sensible starting point, based on that guide:

• Use high for agent loops and coding tasks.
• Use low for routing, tagging and short summaries.
• Keep max for the hard problems only.

Test a small batch first and compare cost per task, not just price per token. Our look at the Claude API request limit change shows the same lesson: small API changes can shift real costs.

A simple checklist before the end of December

You do not need a big project here. Work through these steps in order and you should be done in an afternoon for a small app.

1. Search your code and config for any model name starting with kimi-k2 or moonshot-v1.
2. Swap in kimi-k3.1-pro on a staging key and set reasoning_effort on purpose.
3. Run your usual test prompts and compare answer quality, latency and token use.
4. If you process large offline jobs, try batch mode, which Moonshot says costs 60% of the standard price for kimi-k3.1-pro and kimi-k3.
5. Ship the change, then remove old fallbacks so nothing still points at K2 in January.

One more note from the KissAPI guide: flagship models unlock after a successful top-up of at least $1, so fund the account before launch day.

Frequently Asked Questions

Can I still use Kimi K2 if I already have an account?

Yes, for now. Moonshot says existing users are not affected by the October 31 cutoff, but all K2 models are due to retire by the end of December 2026.

Which model replaces Kimi K2?

Moonshot points developers to a Kimi K3.1 Pro migration guide. It has not published an official model-to-model mapping, so test before you switch production traffic.

Will my Kimi K2 API cost go up after migrating?

Likely yes. The KissAPI guide lists K3.1 Pro at $2 input and $10 output per million tokens, versus $0.95 and $4 for K2.7 Code. Batch mode and a lower reasoning effort can soften the increase.

Do I need to change my code a lot?

Usually not. Per the same guide, the model string changes and the OpenAI SDK works with a different base URL. The main new thing to handle is the reasoning effort setting.

Our take

Do not wait for December. The swap is easy, but the default max reasoning effort and the higher per-token price are the kind of surprises that show up on an invoice, not in a test run. Migrate on a staging key this week, pick your effort levels on purpose, and keep an eye on cost per task. If the new bill stings, look at other models too, since the field is crowded right now.