---
title: "Verify Model and Usage"
description: "Use call logs and five checks to verify the model, token usage, and cost of each call."
headingIds: {"Call Log":"section-1","Five Checks":"section-2","Completing the First Four Checks with a Script":"section-3","thinking Blocks":"section-4","When Inconsistencies Are Found":"section-5","FAQ":"section-6"}
---

<aside className="docs-publication" data-pagefind-ignore>

Documentation maintained by [1ZheAI](https://1zheai.com/en/editorial/#publisher). [Sources and corrections](https://1zheai.com/en/editorial/).

Found an error? [Contact support](mailto:support@1zheai.com) with the page address and a description. Do not send API keys or account credentials.

</aside>
## Call Log

Every call is recorded in the console's [call log](/en/docs/usage-logs/), listing the model, input, output, cache read, and cache write token usage, as well as the cost.

## Five Checks

The following checks compare against the public format of the Anthropic API to detect differences in returned fields and format.

| Check | Expected Result | What to Watch For |
|---|---|---|
| 1. Returned `model` field | Matches the requested model ID | Returns a different model name, or the field is missing |
| 2. Cache read | The same request meeting cache conditions is sent twice within the cache validity period; the second `cache_read_input_tokens` is greater than 0 | Always 0, or no cache fields in `usage` |
| 3. Tool call ID | The `id` of the `tool_use` block starts with `toolu_` | Other formats, or the model does not call tools as required |
| 4. Streaming event order | `message_start`, `content_block_start`, `content_block_delta`, `message_delta`, and `message_stop` appear in sequence | Only OpenAI-format `data:` blocks, or events missing or out of order |
| 5. Real task comparison | Use a fixed task and compare results with the official API or other trusted sources | Results noticeably worse, abnormal refusals, or frequent truncation |

<Warning>

These checks can only detect differences in fields and format; they cannot alone confirm the model actually in use. `model` fields can be rewritten, and formats can be imitated.

</Warning>

## Completing the First Four Checks with a Script

Requires `bash`, `curl`, and `jq`. The script makes five small calls, one of which writes approximately 18,000 tokens to cache, billed at the normal price.

```bash title="verify.sh"
#!/usr/bin/env bash
# 用法：ONEZHE_API_KEY=sk-你的API密钥 bash verify.sh [模型ID]
set -euo pipefail
BASE="https://1zheai.com"
MODEL="${1:-claude-sonnet-5-5}"
H=(-H "x-api-key: $ONEZHE_API_KEY" -H "anthropic-version: 2023-06-01" -H "content-type: application/json")

echo "== 1. model 字段"
curl -s "$BASE/v1/messages" "${H[@]}" \
  -d "$(jq -n --arg m "$MODEL" '{model:$m,max_tokens:16,messages:[{role:"user",content:"ping"}]}')" \
  | jq 'if .error then .error else {id, model, usage} end'

echo "== 2. 缓存：同一请求发两次，看第二次的 cache_read_input_tokens"
LONG=$(printf '这是一段用来测试缓存的固定文本，内容不重要，长度够就行。%.0s' {1..600})
BODY=$(jq -n --arg m "$MODEL" --arg t "$LONG" '{model:$m,max_tokens:16,
  system:[{type:"text",text:$t,cache_control:{type:"ephemeral"}}],
  messages:[{role:"user",content:"ping"}]}')
for i in 1 2; do
  curl -s "$BASE/v1/messages" "${H[@]}" -d "$BODY" | jq -c 'if .error then .error else .usage end'
done

echo "== 3. 工具调用 ID"
curl -s "$BASE/v1/messages" "${H[@]}" \
  -d "$(jq -n --arg m "$MODEL" '{model:$m,max_tokens:256,
    tools:[{name:"get_weather",description:"查询城市天气",
      input_schema:{type:"object",properties:{city:{type:"string"}},required:["city"]}}],
    tool_choice:{type:"auto"},
    messages:[{role:"user",content:"北京今天天气怎么样"}]}')" \
  | jq -r 'if .error then .error.message else (.content[]? | select(.type=="tool_use") | .id) end'

echo "== 4. 流式事件序列"
curl -sN "$BASE/v1/messages" "${H[@]}" \
  -d "$(jq -n --arg m "$MODEL" '{model:$m,max_tokens:32,stream:true,messages:[{role:"user",content:"数到三"}]}')" \
  | grep '^event:' | uniq
```

Check 3 uses `tool_choice: auto`. 5.x models such as Sonnet 5.5 do not support forcing tool specification (`{"type":"tool"}` or `{"type":"any"}`) and will return a 400 `does not support forced tool_choice`.

The output from testing with `claude-sonnet-5-5` on 2026-10-15 is as follows (Check 1 only retains key fields):

```text
== 1. model 字段
{
  "id": "msg_011Cg3Esh8zeEdLeXAPVALKe",
  "model": "claude-sonnet-5-5",
  "usage": {
    "input_tokens": 518,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "output_tokens": 12,
    ...
  }
}
== 2. 缓存：同一请求发两次，看第二次的 cache_read_input_tokens
{"input_tokens":23,"cache_creation_input_tokens":17920,"cache_read_input_tokens":0,"cache_creation":{"ephemeral_5m_input_tokens":17920,"ephemeral_1h_input_tokens":0},"output_tokens":4,"output_tokens_details":{"thinking_tokens":0},"service_tier":"standard","inference_geo":"global"}
{"input_tokens":23,"cache_creation_input_tokens":0,"cache_read_input_tokens":17920,"cache_creation":{"ephemeral_5m_input_tokens":0,"ephemeral_1h_input_tokens":0},"output_tokens":4,"output_tokens_details":{"thinking_tokens":0},"service_tier":"standard","inference_geo":"global"}
== 3. 工具调用 ID
toolu_01UPsdyKX8M9fJTTkDoSKHXA
== 4. 流式事件序列
event: message_start
event: content_block_start
event: ping
event: content_block_delta
event: content_block_stop
event: message_delta
event: message_stop
```

Whether the `ping` event appears and where it appears is not fixed and does not affect the judgment.

## thinking Blocks

When thinking is enabled, the returned `thinking` block includes a `signature` field, which can also serve as a format check. The syntax for 5.x models differs from 4.x; tested on 2026-10-15:

| Model | `thinking` Parameter | Result |
|---|---|---|
| `claude-sonnet-4-6` | `{"type":"enabled","budget_tokens":1024}` | Returns a `thinking` block with thinking text and `signature` |
| `claude-sonnet-5` | `{"type":"enabled","budget_tokens":1024}` or `{"type":"adaptive"}` | Both work |
| `claude-sonnet-5-5` | `{"type":"adaptive"}` | Works; passing `enabled` returns 400, indicating only adaptive is supported |

The `thinking` block of 5.x models contains only `signature` by default, with the thinking text as an empty string. To see the thinking summary, add `"display": "summarized"`:

```json
"thinking": {"type": "adaptive", "display": "summarized"}
```


When using third-party detection websites, it is recommended to create a dedicated API key and delete it after testing.

## When Inconsistencies Are Found

Send an email to support@1zheai.com with the request time, model ID, complete response content (with the API key removed), and the corresponding check item. The request id in the response content or error message can help locate this call.

## FAQ

<AccordionGroup>
<Accordion title="Is asking the model &quot;Who are you?&quot; reliable?">

Unreliable. The model's self-introduction is influenced by the system prompt, and under certain prompts it may answer its own version incorrectly. Checking the returned fields and event format is more informative.

</Accordion>
<Accordion title="Opus is selected — why does Haiku appear in the call logs?">

Claude Code uses the Haiku model for some background tasks, and corresponding records will appear in the call log. You can use the `ANTHROPIC_DEFAULT_HAIKU_MODEL` environment variable to specify the model used for such tasks; see [Claude Code integration](/en/docs/quickstart-claude-code/). If this occurs when not using Claude Code, please send the request time to support@1zheai.com.

</Accordion>
</AccordionGroup>