Summary

Neither parameter setting for maxTokens work with openrouter as a provider, the call falls back to the default hard limit of 4096 output tokens.

Steps to reproduce

1. use the following settings for the model at open router

            "id": "openai/gpt-oss-120b:free",
            "name": "openai/gpt-oss-120b:free (OpenRouter)",
            "api": "openai-completions",
            "reasoning": true,
            "input": [
              "text"
            ],
            "contextWindow": 150000,
            "maxTokens": 40000,
            "compat": {
              "maxTokensField": "max_tokens"
            },
            "headers": {
              "max_tokens": "12000",
              "maxTokensField": "max_tokens"
            }
          },

2. select this model as default on the main agent
3. ask the agent to generate the output of more than 4096 tokens.
4. check the open router log

Expected behavior

The larger number of tokens than 4096 is output

Actual behavior

4096 tokens max whatever the settings are

OpenClaw version

2026.2.23

Operating system

Ubuntu 24.04, node v22

Install method

pnpm build

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response