# Kimi K3 (batch)

As of 12 September 2026, Kimi K3 (batch) costs $3.00 per million input tokens and $15.00 per million output tokens on OpenRouter. Observed on OpenRouter's own published price list on 12 September 2026. This is OpenRouter's resale price, not the model vendor's list price.

Source: OpenRouter (reseller)
Canonical page: https://bertrande.com/en/ai/openrouter/moonshotai--kimi-k3:batch/
Machine formats: https://bertrande.com/en/ai/openrouter/moonshotai--kimi-k3:batch/data.json | https://bertrande.com/en/ai/openrouter/moonshotai--kimi-k3:batch/data.csv
License: CC-BY-4.0, reuse with attribution and a link.

## Parameters

| Parameter | Value | Unit | Observed | Source level | Proof (SHA-256) |
|---|---|---|---|---|---|
| Input price | $3.00 | USD per 1M tokens | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Output price | $15.00 | USD per 1M tokens | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Context window | 1 048 576 | tokens | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Maximum output | 943 718 | tokens | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Agentic index (Artificial Analysis) | 50.6 | index 0-100 | 2026-09-15 | L4 | `48cc0be56adecc1b` |
| Coding index (Artificial Analysis) | 76.2 | index 0-100 | 2026-09-12 | L4 | `134f6a456db7ceff` |
| Intelligence index (Artificial Analysis) | 43.8 | index 0-100 | 2026-09-15 | L4 | `48cc0be56adecc1b` |
| Input modalities | image, text, video | list | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Output modalities | text | list | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Cache read price | $0.300 | USD per 1M tokens | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Listed since | 2026-07-16 | date | 2026-09-12 | L1 | `134f6a456db7ceff` |
| Supported parameters | frequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_p | list | 2026-09-12 | L1 | `134f6a456db7ceff` |

## How this is verified

Every value above was read from the source URL shown on the canonical page, on the date given. We store a SHA-256 fingerprint of the source content at that moment, which is what lets us show later what the source said that day.

No value here was produced, estimated or completed by a language model. Where we have no verified figure, the row is absent rather than filled in.

Method: https://bertrande.com/en/methodology/
Report an error: https://bertrande.com/en/corrections/