Problem

Token usage statistics always show 0 for non-native OpenAI-compatible endpoints like Alibaba Bailian (百炼), even when the API correctly returns usage data.

Root Cause

The issue is in the OpenAI-completions response parser. The code expects input_tokens and output_tokens fields:

// model-selection-_2UcLVem.js line ~103206
input: response.usage?.input_tokens ?? 0,
output: response.usage?.output_tokens ?? 0,
totalTokens: response.usage?.total_tokens ?? 0

However, Bailian API returns prompt_tokens and completion_tokens (standard OpenAI format):

{
  "usage": {
    "prompt_tokens": 11,
    "completion_tokens": 543,
    "total_tokens": 554
  }
}

The normalizeUsage function in OpenClaw correctly handles this fallback pattern:

const rawInput = asFiniteNumber(raw.input ?? raw.inputTokens ?? raw.input_tokens ?? raw.promptTokens ?? raw.prompt_tokens);

But the response parser in the code path for non-streaming responses doesn't use this normalization, causing input_tokens and output_tokens to be undefined, resulting in 0 values.

Additional Issue

For streaming, OpenClaw also forces supportsUsageInStreaming: false for non-native endpoints (CHANGELOG #8714), which may be overly conservative since Bailian supports stream_options: { include_usage: true }.

Steps to Reproduce

1. Configure a non-native OpenAI-compatible endpoint (e.g., Bailian):

{
  "models": {
    "providers": {
      "bailian": {
        "baseUrl": "https://coding.dashscope.aliyuncs.com/v1",
        "api": "openai-completions",
        "models": [{ "id": "qwen3.5-plus", ... }]
      }
    }
  }
}

2. Send messages and check /status - context tokens always show 0.

Actual Behavior

Token usage always shows 0 in /status and session context (e.g., "Context: 0/203k (0%)"), even though the Bailian API correctly returns usage data with prompt_tokens and completion_tokens fields.

Expected Behavior

Token usage should be correctly parsed from prompt_tokens/completion_tokens fallback fields, matching the behavior of the normalizeUsage function.

Environment

• OpenClaw: 2026.3.12
• Provider: Alibaba Bailian (百炼)
• API: https://coding.dashscope.aliyuncs.com/v1