# Kimi K3 vs Kimi K3.1

As of November 21, 2026, Kimi K3.1 costs 47% more per input token than Kimi K3 ($4.40 vs $3.00 per 1M) and has a 869K-token context window versus 1M.

- Page: https://models.litellm.ai/compare/kimi-k3-vs-kimi-k3.1
- Kimi K3: https://models.litellm.ai/models/kimi-k3
- Kimi K3.1: https://models.litellm.ai/models/kimi-k3.1

## Side by side

| | Kimi K3 | Kimi K3.1 |
|---|---|---|
| Input, per 1M tokens | $3.00 | $4.40 |
| Output, per 1M tokens | $15.00 | $22.00 |
| Cache read, per 1M tokens | $0.30 | $0.22 |
| Cache write, per 1M tokens | $3.00 | $5.50 |
| Context window | 1,048,576 tokens | 868,928 tokens |
| Max output | 1,048,576 tokens | 131,072 tokens |
| Image input | Yes | Yes |
| Tool calling | Yes | Yes |
| Structured output (JSON) | Yes | Yes |
| Prompt caching | Yes | Yes |
| Reasoning | Yes | Yes |
| Providers on LiteLLM | 21 | 1 |
| Price shown is from | Moonshot AI | Amazon Bedrock |
| Added to LiteLLM | August 16, 2026 | November 7, 2026 |

Headline prices in USD per 1M tokens, from LiteLLM's model price file.

## Cheapest provider for each

- Kimi K3: Sail at $2.50 input per 1M tokens (`sail/moonshotai/Kimi-K3`)
- Kimi K3.1: Amazon Bedrock at $4.40 input per 1M tokens (`moonshotai.kimi-k3.1`)

## When to pick which

- Kimi K3: Costs 32% less per input token: $3.00 vs $4.40 per 1M.
- Kimi K3: Larger context window: 1,048,576 vs 868,928 tokens.

## FAQ

### Which is cheaper, Kimi K3 or Kimi K3.1?

Kimi K3 is cheaper at 3 input tokens per output token: $6.00 per 1M tokens blended ($3.00 input, $15.00 output), versus $8.80 ($4.40 and $22.00) for Kimi K3.1, as of November 21, 2026.

### Which has the larger context window, Kimi K3 or Kimi K3.1?

Kimi K3 has the larger context window: 1,048,576 tokens, versus 868,928 for Kimi K3.1.

### Can I switch from Kimi K3 to Kimi K3.1 without changing my code?

Yes. Through LiteLLM both use the OpenAI format, so switching means changing the model name from "moonshot/kimi-k3" to "moonshotai.kimi-k3.1", or pointing a LiteLLM gateway model alias at the other model. You can also use one as a fallback for the other.
