How do I use DeepSeek in OpenClaw?
Add a standardcompute provider under models.providers in ~/.openclaw/openclaw.json (baseUrl https://api.stdcmpt.com/v1) with deepseek/deepseek-v4.1-flash in its models list, set agents.defaults.model.primary to standardcompute/deepseek/deepseek-v4.1-flash and run openclaw config validate: OpenClaw then runs on DeepSeek V4.1 Flash on a plan from $19 a month that returns HTTP 402 to new requests once its compute budget is used.
{
"models": {
"mode": "merge",
"providers": {
"standardcompute": {
"baseUrl": "https://api.stdcmpt.com/v1",
"apiKey": "${STANDARDCOMPUTE_API_KEY}",
"api": "openai-responses",
"models": [
{
"id": "deepseek/deepseek-v4.1-flash",
"name": "DeepSeek V4.1 Flash",
"reasoning": true,
"input": ["text", "image"],
"contextWindow": 400000,
"maxTokens": 32000
},
{
"id": "deepseek/deepseek-v4-pro",
"name": "DeepSeek V4 Pro",
"reasoning": true,
"input": ["text"],
"contextWindow": 400000,
"maxTokens": 32000
}
]
}
}
},
"agents": {
"defaults": {
"model": {
"primary": "standardcompute/deepseek/deepseek-v4.1-flash"
}
}
}
}
Model refs split on the first slash, so standardcompute/deepseek/deepseek-v4.1-flash means provider standardcompute, model deepseek/deepseek-v4.1-flash. Merge the blocks into your existing file.
Validate before you restart
OpenClaw accepts only configs that match its schema: an unknown key, a wrong type or an invalid value stops the Gateway from starting. openclaw config validate checks the file without starting anything, and openclaw doctor --fix drops unknown keys after a backup. A model referenced in agents.defaults also needs its own entry in models.providers.standardcompute.models, or OpenClaw cannot resolve it.
What works with DeepSeek in OpenClaw
| OpenClaw feature | On DeepSeek through Standard Compute |
|---|---|
| Images | V4.1 Flash is marked text and image, so OpenClaw injects attachments. V4 Pro is text-only |
| Thinking | With reasoning: true, /think levels reach the model as a reasoning effort; /think off omits it |
| Model refs | Split on the first slash: standardcompute/deepseek/deepseek-v4.1-flash |
| Switching | /model standardcompute/deepseek/deepseek-v4-pro for one session |
From OpenClaw's custom-provider, configuration, models and thinking docs, checked October 14, 2026.
Heartbeats are where an OpenClaw day goes
OpenClaw runs a heartbeat every 30 minutes by default, and each one is a full agent turn. Its docs suggest isolatedSession: true (about 2K to 5K tokens per run instead of about 100K), lightContext: true or a cheaper heartbeat model.
| Per day | V4.1 Flash, peak | Off-peak |
|---|---|---|
| Typical day | $0.42 | $0.21 |
| Busy day | $4.31 | $2.16 |
DeepSeek list prices per million tokens for V4.1 Flash: $0.30 input, $0.006 cache hit, $1.20 output at peak, half off-peak. Source. The Agent Burn Index measures coding agents such as OpenCode, Cline and Kilo; OpenClaw is not broken out, so these figures use its median and 90th-percentile coding-agent day (September 13 to October 12, 2026) as a stand-in. Priced at the list rates above; your own days will differ.
The catch
OpenClaw's DeepSeek plugin (openclaw plugins install @openclaw/deepseek-provider, DEEPSEEK_API_KEY) runs deepseek/deepseek-flash at DeepSeek's prices; onboarding sets V4 Pro as its default. OpenClaw's own security docs also warn that cheaper model tiers are more open to prompt injection and recommend the strongest model for tool-enabled bots, so think twice before running an unattended OpenClaw on a budget model alone.
Where Standard Compute fits
A monthly cap for a Gateway that never sleeps, and a stronger model on the same key: add Claude Sonnet 5.5 to the models list and make it the fallback or the primary for tool-heavy agents.
FAQ
Why does OpenClaw send my key as the literal text STANDARDCOMPUTE_API_KEY?
apiKey needs the "${STANDARDCOMPUTE_API_KEY}" form. A bare uppercase name is sent as the key itself, except for a few names OpenClaw knows.
Is deepseek/deepseek-chat still a valid OpenClaw model?
No. DeepSeek retired deepseek-chat and deepseek-reasoner on July 24, 2026; OpenClaw's DeepSeek plugin now uses deepseek/deepseek-flash and deepseek/deepseek-v4-pro.
Can I use the launcher instead of editing openclaw.json?
Yes: standardcompute launch openclaw opens OpenClaw in a separate profile on smart routing and leaves your config alone. Edit openclaw.json when you want DeepSeek pinned in your own setup.