# Moonshot: k3

- Model ID: `k3`
- Provider: Moonshot
- Web version: https://www.moleapi.com/en/models/moonshot/kimi-k3
- Content status: alias

## Model introduction

Kimi K3 is Moonshot AI's 2.8T-parameter native multimodal reasoning model with a 1M context window, built for long-horizon coding, knowledge work, and deep reasoning.

## Model capabilities

- OpenAI compatible
- Responses API
- Anthropic compatible
- Gemini compatible
- Reasoning

### Verified specifications

- Official positioning: Long-horizon coding, knowledge work, and deep reasoning
- Context window: 1,000,000 tokens
- Model size: 2.8T parameters, 16 / 896 experts active
- Input modalities: Text, image

## Model pricing and access

Pricing is supplied dynamically by the MoleAPI console API.

### Standard (default, x1)

- Input: $3 / 1M tokens
- Output: $15 / 1M tokens
- Cache read: $0.3 / 1M tokens

### Discount (discount, x0.8)

- Input: $2.4 / 1M tokens
- Output: $12 / 1M tokens
- Cache read: $0.24 / 1M tokens

### Relay (relay, x0.3)

- Input: $0.9 / 1M tokens
- Output: $4.5 / 1M tokens
- Cache read: $0.09 / 1M tokens

### Access protocols

| Protocol | Method | Endpoint |
| --- | --- | --- |
| openai | POST | /v1/chat/completions |
| openai-response | POST | /v1/responses |
| anthropic | POST | /v1/messages |
| gemini | POST | /v1beta/models/{model}:generateContent |

- Live pricing source: https://home.moleapi.com/api/pricing

## About k3

Kimi K3 is Moonshot AI's 2.8T-parameter native multimodal reasoning model with a 1M context window, built for long-horizon coding, knowledge work, and deep reasoning.

### Best for

Large repositories, long-running agents, research-heavy knowledge work, and coding or creation loops that use visual feedback.

### Core strengths

- A 2.8T architecture, 1M context, and native vision support unusually long agent workflows.
- Its independent composite score is very strong, especially for long-horizon coding and knowledge work.
- An open-weight path and OpenAI-compatible API support migration and future self-hosting.

### Limitations

- It currently defaults to maximum reasoning effort; independent testing shows slow output and high token use.
- Moonshot notes that the model can be overly proactive, so bounded agents need explicit system constraints.
- Switching models mid-session or dropping thinking history can make quality unstable.

### Selection and production evaluation

Start a k3 evaluation by mapping its official positioning to real work: Large repositories, long-running agents, research-heavy knowledge work, and coding or creation loops that use visual feedback. The first pass should exercise both its main strength, "A 2.8T architecture, 1M context, and native vision support unusually long agent workflows.", and its known limitation, "It currently defaults to maximum reasoning effort; independent testing shows slow output and high token use.", instead of relying on a single subjective general-chat comparison.

For access, MoleAPI currently lists openai, openai-response, anthropic, gemini protocols and the Standard, Discount, Relay billing groups for this Moonshot model; the default price summary is Input $3 / 1M tokens · Output $15 / 1M tokens. Pricing, protocols, and groups come from the live catalog, so production planning should still price representative requests using real context length, output size, and cache-hit assumptions.

No independent ranking is shown unless it matches this exact model ID and reasoning profile, so nearby variants are not used as a proxy. Before launch, pin the model ID, prompt, and sample set, then compare task accuracy, structured-output validity, tool-call success, and timeout rates under the same conditions.

The model material on this page was last checked on 2026-07-23. When upstream model cards, context limits, or tool support change, update the cited bilingual facts before changing the recommendation; live MoleAPI price changes remain separate and update from the catalog automatically.

### Sources

- [Moonshot AI Kimi K3 technical blog](https://www.kimi.com/blog/kimi-k3) — Moonshot AI
- Last verified: 2026-07-23

## Code examples

### cURL

```bash
curl https://api.moleapi.com/v1/responses \
  -H "Authorization: Bearer $MOLEAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"k3","input":"Explain this problem step by step."}'
```

### Python

```python
from openai import OpenAI

client = OpenAI(base_url="https://api.moleapi.com/v1", api_key="YOUR_MOLEAPI_API_KEY")
response = client.responses.create(model="k3", input="Explain this problem step by step.")
print(response.output_text)
```

### TypeScript

```typescript
import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.moleapi.com/v1", apiKey: process.env.MOLEAPI_API_KEY });
const response = await client.responses.create({ model: "k3", input: "Explain this problem step by step." });
console.log(response.output_text);
```

## More related models

- [kimi-k2.6](https://www.moleapi.com/en/models/moonshot/kimi-k2.6) — Moonshot
- [kimi-k3](https://www.moleapi.com/en/models/moonshot/kimi-k3) — Moonshot
- [kimi-k2.7-code](https://www.moleapi.com/en/models/moonshot/kimi-k2.7-code) — Moonshot
- [gpt-5.6-luna](https://www.moleapi.com/en/models/openai/gpt-5.6-luna) — OpenAI
- [gpt-5.6-sol](https://www.moleapi.com/en/models/openai/gpt-5.6-sol) — OpenAI
- [gpt-5.6-terra](https://www.moleapi.com/en/models/openai/gpt-5.6-terra) — OpenAI

## Frequently asked questions

### How is k3 priced?

Prices are read from the MoleAPI console API and update with model prices, context tiers, and account groups.

### How can I access k3?

The catalog currently lists openai, openai-response, anthropic, gemini.

### How do I switch an existing project to k3?

Keep the MoleAPI API address and key, replace the model parameter, and check protocol-specific parameters.

### Where do the k3 details come from?

Capabilities and limitations are checked against Moonshot AI and the other cited pages; pricing and protocols come from the MoleAPI console.