2.0 KiB

Raw Blame History

K2.5 Emergency Model Switch

Status: PREPARED - NOT ACTIVE Purpose: Backup if Anthropic credits run out

K2.5 Model Details

Provider: Fireworks.ai
Model ID: accounts/fireworks/models/kimi-k2p5 (verified available)
Also available: kimi-k2-instruct-0905, kimi-k2-thinking
Context: 128k tokens
Max output: 8192 tokens
Cost: ~$0.40/M input, ~$1.60/M output (10% of Opus)

To Activate (Emergency Only)

Option 1: Per-session override (temporary)

/model fireworks/accounts/fireworks/models/kimi-k2p5

Option 2: Set as default for main agent

Add to agents.list[0] (main agent):

{
  "id": "main",
  "model": "fireworks/accounts/fireworks/models/kimi-k2p5",
  "identity": {
    "name": "James",
    "emoji": "⚡"
  }
}

Option 3: Full config patch (adds K2.5 to providers + sets default)

Run this command:

openclaw gateway config.patch --raw '{
  "models": {
    "providers": {
      "fireworks": {
        "models": [
          {
            "id": "accounts/fireworks/models/kimi-k2p5",
            "name": "Kimi K2.5",
            "reasoning": false,
            "input": ["text"],
            "cost": {"input": 0.40, "output": 1.60, "cacheRead": 0, "cacheWrite": 0},
            "contextWindow": 128000,
            "maxTokens": 8192
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": "fireworks/accounts/fireworks/models/kimi-k2p5"
    }
  }
}'

K2.5 Limitations vs Opus

Smaller context (128k vs 200k)
Less nuanced reasoning
May struggle with complex multi-step tasks
Good for: Straightforward coding, simple Q&A, routine tasks
Not ideal for: Complex judgment calls, nuanced conversation

Rollback to Anthropic

openclaw gateway config.patch --raw '{
  "agents": {
    "defaults": {
      "model": "anthropic/claude-opus-4-5"
    }
  }
}'

Or per-session: /model anthropic/claude-opus-4-5

Fireworks API Key

Already configured in env: FIREWORKS_API_KEY

2.0 KiB Raw Blame History