# Kimi K2.6

> Moonshot's newest. Agentic + vision + reasoning. 262K context.

Human version: https://kymaapi.com/models/kimi-k2.6
Live JSON: `GET https://kymaapi.com/v1/models` (no auth required)

## Facts

- **Model ID**: `kimi-k2.6` — pass this as `model` in the request body
- **Creator**: Moonshot AI
- **Released**: 2026-04-20
- **Context window**: 262K tokens
- **Max output**: 33K tokens per response — a hard ceiling, not a default
- **Price**: $0.776 in / $3.622 out per 1M
- **Capabilities**: tools, vision, reasoning
- **Aliases**: `agent`, `best-agent` resolve to this model

## Call it

```bash
curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k2.6", "messages": [{"role": "user", "content": "Hello"}]}'
```

## Positioning

Kimi K2.6 is a frontier model optimized for agentic workflows, visual reasoning, and long-context tasks. Route to this model when your application requires multi-step tool use, image analysis, or structured outputs over raw throughput.

## About

Created by Moonshot, this model accepts text and image inputs with a 262K token context window and a 32K token output limit. It natively supports tool calling, explicit reasoning steps, and structured JSON outputs.

On Kyma, it runs through an OpenAI-compatible endpoint with automatic request failover. The platform reports exact request costs in the usage.cost field and identifies the routed model via the X-Kyma-Model response header. While Kyma bills cached prefixes at 10% of the input rate, this model does not support caching, so all input is billed at the standard rate.

It operates in the premium cost tier with medium generation speed. Early platform traffic shows a 100% success rate, but the 32K output cap means it is not suited for generating extremely long documents in a single pass.

## Use cases

- **Multi-step agent routing** — Routes tool calls and maintains state across extended conversational turns.
- **Image and text analysis** — Extracts and analyzes visual data from uploaded images alongside text prompts.
- **Long-context code review** — Processes large codebases within the 262K window to identify architectural patterns.
- **Structured output generation** — Outputs formatted payloads reliably using built-in structured output support.

## Not ideal for

Do not use this model for high-throughput, low-latency tasks or when you need prompt caching to discount repeated input costs.

## Pick something else when

- You need faster generation at a lower cost tier → `deepseek-v4-flash`, `gemma-4-31b`
- You require prompt caching to discount repeated prefixes → `qwen-3.7-max`, `deepseek-v3`
- You need pure text generation without vision capabilities → `qwen-3.7-plus`, `minimax-m2.5`

## See also

- All models: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md
- Other models by Moonshot AI: https://kymaapi.com/models?q=Moonshot%20AI