K3's partial mode is no longer available

Kimi K3
1

Happened at yesterday’s afternoon. While I tried to use K3’s partial mode with “partial: true”, model can only reply 429 failure:The engine is currently overloaded. I’ve tried every methods including api, kimicode and openrouter. Are kimi going to fix this?

By the way, the rest of the model like k2.8, k3.1 is working fine with partial mode.

2

Thanks for your support!

The "partial": true parameter is actually model-agnostic—generally speaking, it should be supported across all models.

A 429 error isn’t related to Partial Mode itself. In our service, 429 covers a group of scenarios (including engine_overloaded_error and rate_limit_reached_error). Since other models like k2.8 and k3.1 are working fine for you, your account quota is clearly fine—it’s most likely triggered by temporary server-side capacity pressure on that specific model node, or hitting concurrency/rate limits (RPM/TPM).

Could you share a simplified, sanitized curl request along with the exact error response payload (specifically the error type and message) so I can try reproducing it on my end?

For example, something like this with the endpoint, model, and parameters you’re using:

curl https://api.moonshot.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $MOONSHOT_API_KEY" \
  -d '{
    "model": "YOUR_MODEL_ID",
    "messages": [
      {"role": "user", "content": "Hello"},
      {"role": "assistant", "content": "Hi, ", "partial": true}
    ]
  }'
3

I called the api like this:
curl https://api.kimi.com/coding/v1/chat/completions
-H “Content-Type: application/json”
-H “Authorization: Bearer $KIMI_API_KEY”
-d ‘{
“model”: “k3”,
“max_tokens”: 64,
“stream”: false,
“messages”: [
{
“role”: “user”,
“content”: “Hello”
},
{
“role”: “assistant”,
“content”: "Hi, ",
“partial”: true
}
]
}’

and it replied:

{“error”:{“message”:“The engine is currently overloaded, please try again later”,“type”:“engine_overloaded_error”}}

I didn’t fill my actual api at here but I already show the error.

4

By the way I don’t think it’s a account relating failure since my openrouter calling also received “provider error” like: OpenRouter endpoint failed with status 429: Provider returned error. It mostly comes from Moonshot AI provider while other provider such as Fireworks and BaseTen work just fine with parameter “partial: true”.

5

Thanks for following up with the detailed payload.

We checked the backend infrastructure for K3 and identified the cause. Partial Mode is still supported, but the feature flag was not consistently enabled across all serving clusters. Following recent traffic and capacity adjustments, requests with "partial": true had very limited routing capacity, which triggered the engine_overloaded_error you observed.

We have aligned the configuration across the affected clusters, and the fix has been live for a while now. Please retry your original request when convenient. If you still encounter failures, please include the timestamp and request ID, if available, and we will investigate further.

Thanks again for the detailed report.

1 Like
6

It worked fine with me now, thanks.