Use Kimi K3 in goose
goose runs Kimi K3 on OceanToken with three settings: a base URL, your key and the model id. Tested on October 19, 2026 with goose CLI 1.54.0.
Settings
| Setting | Value |
|---|---|
| Provider | OceanToken (custom provider file) |
| Provider file | https://docs.oceantoken.ai/goose/oceantoken.json |
| OCEANTOKEN_API_KEY | your OceanToken API key |
| Model | moonshotai/kimi-k3 |
From the command line
mkdir -p ~/.config/goose/custom_providers
curl -fsSL -o ~/.config/goose/custom_providers/oceantoken.json https://docs.oceantoken.ai/goose/oceantoken.json
export OCEANTOKEN_API_KEY=sk-...
goose session --provider oceantoken --model moonshotai/kimi-k3What it costs
An agent resends a long prompt on every turn, so the cached input price matters as much as the input price. Kimi K3 caches the repeated part of the prompt on its own: from the second turn of a task it is billed at the cached input price.
About these prices
Frequently asked questions
Does Kimi K3 work with goose’s tools?
Yes. It was tested on October 19, 2026 with goose CLI 1.54.0: writing and running a small program with its file and shell tools.
How much does Kimi K3 cost in goose?
Every request goose makes is billed at the model’s price: $3.00 / $15.00 per 1M input / output tokens. The repeated part of a long prompt is billed at $0.30 per 1M cached input tokens. Each request’s charge is in the console under Logs.
Where do I put my OceanToken key?
Create a key in the OceanToken console and give it a credit limit: an agent sends many requests, and the limit caps what this key can spend. Keep it in an environment variable, never in a file in your repository.
Other models tested with goose
DeepSeek V4 Pro · DeepSeek V4 Flash 0423 · MiniMax M3
Keep exploring
Kimi K3 vs DeepSeek V4 Pro · Kimi K3 vs Claude Sonnet 5.5 · GLM 5.3 vs Kimi K3 · Kimi K3 vs Claude Opus 5.5 · Cheaper alternatives to Kimi K3