#Ollama cloud models lack think options

1 messages · Page 1 of 1 (latest)

stuck spoke
#

I'm trying to change the think mode via the telegram /think command for model deepseek-v4-flash:cloud but the only option is "off". This is my models block in openclaw.json:

  "models": {
    "mode": "merge",
    "providers": {
      "ollama": {
        "baseUrl": "http://127.0.0.1:11434",
        "api": "ollama",
        "apiKey": "OLLAMA_API_KEY",
        "models": [
          {
            "id": "kimi-k2.5:cloud",
            "name": "kimi-k2.5:cloud",
            "reasoning": false,
            "input": [
              "text",
              "image"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 128000,
            "maxTokens": 8192,
            "compat": {
              "supportsTools": false
            }
          },
          {
            "id": "minimax-m2.7:cloud",
            "name": "minimax-m2.7:cloud",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 128000,
            "maxTokens": 8192
          },
          {
            "id": "glm-5.1:cloud",
            "name": "glm-5.1:cloud",
            "reasoning": true,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 202752,
            "maxTokens": 8192,
            "compat": {
              "supportsTools": true
            }
          },
          {
            "id": "qwen3-embedding:0.6b",
            "name": "qwen3-embedding:0.6b",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 32768,
            "maxTokens": 8192,
            "compat": {
              "supportsTools": false
            }
          },
          {
            "id": "kimi-k2.6:cloud",
            "name": "kimi-k2.6:cloud",
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 262144,
            "maxTokens": 8192,
            "compat": {
              "supportsTools": true
            }
          },
          // ... some more models after this, deepseek isn't listed though
        ]
      }
    }
  }

And here's my agents.defaults model and models blocks

      "model": {
        "primary": "ollama/kimi-k2.6:cloud"
      },
      "models": {
        "ollama/kimi-k2.6:cloud": {},
        "ollama/glm-5.1:cloud": {},
        "ollama/gemma4:31b-cloud": {},
        "ollama/deepseek-v4-flash:cloud": {}
      },

Response: Thinking level "high" is not supported for ollama/deepseek-v4-flash:cloud. Use one of: off.

fringe estuaryBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

dire brambleBOT
#

First, can you share openclaw --version?

That said, the issue is likely here: deepseek-v4-flash:cloud is only in agents.defaults.models, not in models.providers.ollama.models with reasoning: true.

For Ollama, /think options come from the resolved model metadata. If OpenClaw sees that model as non-reasoning, it only exposes off.

Add DeepSeek to the provider model list:

{
  id: "deepseek-v4-flash:cloud",
  name: "deepseek-v4-flash:cloud",
  reasoning: true,
  input: ["text"],
  contextWindow: 128000,
  maxTokens: 8192
# 
}

Keep this too if you want it selectable:

"agents": {
  "defaults": {
    "models": {
      "ollama/deepseek-v4-flash:cloud": {}
    }
  }
}

Then restart/reload the gateway and try:

stuck spoke
#

OpenClaw version 2026.4.26. I added that and restarted the gateway and observed this:
[29/04/2026 4:13 PM] 81442: /model ollama/deepseek-v4-flash:cloud
[29/04/2026 4:14 PM] Claw: Model set to ollama/deepseek-v4-flash:cloud.
[29/04/2026 4:14 PM] 81442: /think
[29/04/2026 4:14 PM] Claw: Choose level for /think.
Options: off.

Still says off.

dire brambleBOT
#

Ah, thanks — that confirms it. You’re likely hitting a 2026.4.26 bug, not a config mistake.

The relevant fix is already in current main/changelog: Ollama /think validation was changed to use live/catalog reasoning metadata so thinking-capable Ollama models don’t get stuck with only off. That fix appears after 2026.4.26, so adding reasoning: true may not affect the /think menu yet.

Workaround until the next release: force the native Ollama think param on the model entry:

{
  id: "deepseek-v4-flash:cloud",
  name: "deepseek-v4-flash:cloud",
  reasoning: true,
  input: ["text"],
  contextWindow: 128000,
  maxTokens: 8192,
  params: {
    thinking: "high"
# 
  }
}

or thinking: "medium" / "low".

That won’t make /think show the choices, but it should make the request send Ollama’s top-level think value. The proper /think menu fix should come with the next OpenClaw release. Docs note that /think options are provider-profile driven, and explicit Ollama configs disable auto-discovery, so this is exactly the path affected here. (docs.openclaw.ai)

stuck spoke
#

ok when will 4.27 release