Jatevo — LLM Inference for OpenClaw

Access 3 flagship LLMs via a single API key. OpenAI-compatible, production-grade.

Jatevo.ai provides inference-as-a-service for open-source LLMs, hosted in Asia with international endpoints.

All models are OpenAI-compatible and production-tested.

About Qwen 3.5 Plus

Qwen 3.5 Plus is the hosted version of Alibaba’s flagship Qwen3.5-397B-A17B — the same open-weight model, served via Alibaba Cloud with enhanced features:

• 397 billion parameters, 17B active per token via Sparse MoE + Gated DeltaNet hybrid architecture
• 1M token context (vs 256K for the open-weight version)
• Native multimodal — text + image understanding
• 201 languages supported
• Apache 2.0 open-weight model, released Feb 16, 2026

AI Benchy Global Ranking (March 2026)

Qwen 3.5 Plus ranks #8 globally on AI Benchy, beating GPT-5.4, Claude Sonnet 4.6, and Claude Opus 4.6:

Source: aibenchy.com — Updated March 20, 2026

Pricing

🎁 Campaign: We’re giving away 190M free tokens to support Indonesia’s developer community. Claim yours →

Setup

Option 1: Web Setup Guide (recommended)

Go to jatevo.ai/private-ai/openclaw-setup — follow the step-by-step guide. Your API key will be pre-filled after you sign in. No scripts, no env vars, no crash risk.

Option 2: Manual Setup

1. Get your API key at jatevo.ai — sign in and go to the API section
2. Add to your OpenClaw config (~/.openclaw/openclaw.json):

Add this under models.providers — replace jk_your_key_here with your actual key:

{
  "models": {
    "mode": "merge",
    "providers": {
      "jatevo": {
        "baseUrl": "https://jatevo.id/api/open/v1/inference",
        "apiKey": "jk_your_key_here",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen3.5-plus",
            "name": "Qwen 3.5 Plus",
            "reasoning": true,
            "input": ["text", "image"],
            "cost": { "input": 0.0008, "output": 0.002, "cacheRead": 0, "cacheWrite": 0 },
            "contextWindow": 1000000,
            "maxTokens": 65536
          },
          {
            "id": "kimi-k2.5",
            "name": "Kimi K2.5",
            "reasoning": true,
            "input": ["text", "image"],
            "cost": { "input": 0.0006, "output": 0.003, "cacheRead": 0, "cacheWrite": 0 },
            "contextWindow": 262144,
            "maxTokens": 32768
          },
          {
            "id": "glm-4.7",
            "name": "GLM 4.7",
            "reasoning": true,
            "input": ["text"],
            "cost": { "input": 0.0001, "output": 0.0002, "cacheRead": 0, "cacheWrite": 0 },
            "contextWindow": 128000,
            "maxTokens": 40000
          }
        ]
      }
    }
  }
}

⚠️ Important: Put your actual jk_... key directly in the apiKey field. Do NOT use ${JATEVO_API_KEY} — that requires an env var and will crash your gateway if it’s not set.

3. Restart gateway:

openclaw gateway restart

That’s it. Done.

Option 3: Setup script

git clone https://github.com/lucacadalora/jatevo-openclaw ~/.openclaw/skills/jatevo
bash ~/.openclaw/skills/jatevo/setup.sh
openclaw gateway restart

The script will ask for your API key and embed it directly in your config.

Speed Benchmarks

Tested March 21, 2026 on production workloads via Jatevo API:

Qwen 3.5 Plus │ Short: 38 tok/s │ Code: 59 tok/s │ Long: 56 tok/s │ Avg: 51.0 tok/s
Kimi K2.5     │ Short: 31 tok/s │ Code: 33 tok/s │ Long: 37 tok/s │ Avg: 33.6 tok/s
GLM 4.7       │ Short: 285 tok/s│ Code: 531 tok/s│ Long: 612 tok/s│ Avg: 475.9 tok/s 🔥

When to Use Which

• Need the smartest model? → jatevo/qwen3.5-plus (397B params, 1M context, vision, #8 globally)
• Need deep reasoning? → jatevo/kimi-k2.5 (strongest chain-of-thought)
• Need it NOW? → jatevo/glm-4.7 (476 tok/s — 9x faster than Qwen)
• Huge document? → jatevo/qwen3.5-plus (1M token context window)
• Analyze images? → jatevo/qwen3.5-plus or jatevo/kimi-k2.5
• Building AI agents? → jatevo/qwen3.5-plus (72.2 BFCL-V4 tool use)

Switching Models

After setup, switch models anytime in OpenClaw:

/model jatevo/qwen3.5-plus
/model jatevo/kimi-k2.5
/model jatevo/glm-4.7

Uninstall

# Remove from config
# Edit ~/.openclaw/openclaw.json and remove the "jatevo" provider block
openclaw gateway restart

Rate Limits

• ~1,200 prompts per 5 hours (~4 req/min sustained)
• Fully parallel — tested with 32 concurrent requests
• No per-model limits

Structure

├── README.md    ← You are here
├── SKILL.md     ← OpenClaw skill definition + detailed model specs
├── setup.sh     ← Interactive setup script
└── LICENSE      ← MIT

Support

• Website: jatevo.ai
• Issues: GitHub Issues
• Community: OpenClaw Discord

License

MIT — see LICENSE