openclaw.json

// In agents.models:
"nvidia/moonshotai/kimi-k2.5": {
  "alias": "kimi"
}

// In the root
"models": {
  "mode": "merge",
  "providers": {
    "nvidia": {
      "baseUrl": "https://integrate.api.nvidia.com/v1",
      "apiKey": "<API_KEY>",
      "api": "openai-completions",
      "models": [
        {
          "id": "moonshotai/kimi-k2.5",
          "name": "Kimi K2.5",
          "reasoning": true,
          "input": [
            "text",
            "image"
          ],
          "cost": {
            "input": 0,
            "output": 0,
            "cacheRead": 0,
            "cacheWrite": 0
          },
          "contextWindow": 256000,
          "maxTokens": 8192
        }
      ]
    }
  }
}

Yes! Sry. I also messed up the output: 262144. Free with Ollama free account. I am not sure about the limit cos I didn't follow tokens, but I hit the limit around the same on bith nvidia and ollama. But ollama is much faster. At least for me

Here are the essential facts regarding Kimi-K2.5:cloud quotas on the Ollama Cloud Free Tier:

• Token Context: Supports up to 256,000 (256K) tokens for input.
• Generation Limit: Output is capped at approximately 16,384 tokens per response.
• Usage Model: Operates on a percentage-based system (usage credits) rather than a fixed token count.
• Hourly Quota: Roughly 250,000 input tokens per hour (subject to system load).
• Request Rate: Approximately 135 requests every 5 hours for typical chat usage.
• Resource Intensity: As a 1-trillion parameter model, Kimi-K2.5 consumes your usage percentage faster than smaller models (like Llama 3).
• Monitoring: Live usage tracking is available in your Ollama Dashboard under account settings.

My config: https://github.com/rainman74/OpenClaw-Configs/blob/main/.openclaw/Windows/openclaw.json