EN简中

ModelDial model check

Model check

Has ChatGPT or Codex become less capable, or has your API provider switched models? Use a fingerprint check to help verify the model you are using.

Two fingerprint probesNo installationCompared in your browser
How it works

Fingerprint coverage

14 model families · 64 models

Models in the database

Browse the models in each family. You can also test unlisted models; results will show the closest matches in the database.

Models: 64

ClaudeModels: 12

  • claude-haiku-4.5
  • claude-sonnet-4.6
  • claude-sonnet-5
  • claude-opus-4.6
  • claude-opus-4.7
  • claude-opus-4.8
  • claude-opus-5
  • claude-fable-5.1
  • claude-fable-5
  • claude-opus-5.5
  • claude-sonnet-5.5
  • claude-haiku-5.5

GPTModels: 10

  • gpt-4o
  • gpt-6-luna
  • gpt-6-sol
  • gpt-6-astra/6.1-sol
  • gpt-5.4
  • gpt-5.5
  • gpt-5.6-sol
  • gpt-5.6-terra
  • gpt-5.6-luna
  • gpt-6.1-terra

GeminiModels: 8

  • gemini-2.5-pro
  • gemini-3.1-pro-preview
  • gemini-3.5-flash
  • gemini-3.6-flash
  • gemini-3.7-flash
  • gemini-3.8-flash
  • gemini-4-argon
  • gemini-3.5-flash-lite

DeepSeekModels: 7

  • deepseek-v4.1-flash
  • deepseek-v4-flash
  • deepseek-v4-pro-0813
  • deepseek-v4-pro
  • deepseek-v4-flash-0731
  • deepseek-v3.2
  • deepseek-v4.1-flash-1017

GrokModels: 3

  • grok-4.5
  • grok-4.6
  • grok-4.7

QwenModels: 4

  • qwen3.8-max-0902
  • qwen3.8-27b
  • qwen3.8-omni-flash
  • qwen3.8-flash

GLMModels: 4

  • glm-5.2
  • glm-5.3
  • glm-5.3-flashx
  • glm-5.3-flash

KimiModels: 3

  • kimi-k3
  • kimi-k2.8-preview
  • kimi-k3.1-pro

MiniMaxModels: 1

  • minimax-m3.1-flash

MiMoModels: 4

  • mimo-v2.5
  • mimo-v2.5-pro
  • mimo-v2.6-flash
  • mimo-v2.6-pro

SeedModels: 3

  • seed-2.0-mini
  • seed-2.0-lite
  • seed-2.1-turbo

StepModels: 1

  • step-5-preview

HunyuanModels: 1

  • hy4-preview

MuseModels: 3

  • muse-spark-1.2
  • muse-spark-1.3
  • muse-wonder

Start a check

Enter your API URL and key, choose a model, then start the check.

Your key passes through the ModelDial relay to your API. It is not written to a database, cache or logs. The input is cleared when a check starts.

We record the check method, model name and matching result to improve detection. API keys, original answers and API addresses are not stored.

  1. Connect your API

    API service

    Click an official service to fill in its URL. For a third-party service, keep Custom and paste the API URL your provider gave you. The protocol is detected automatically.

    Paste the URL exactly as your provider gives it, such as …/v1 or …/anthropic. A bare domain uses /v1.

    Use a temporary API key for testing and revoke it afterward.

  2. Select a model to test

    A reference will be suggested after you select a model.

    Model to test

    Fetching the list only reads the catalog. It does not generate answers.

    Reference · ModelDial fingerprint database

    After you select a model to test, we will suggest a reference from the fingerprint database.

  3. Choose probes and start

    Advanced settings: connection, protocol and reasoning
    Connection
    API protocol

    Auto-detect picks a protocol from the URL and model, and moves to the next one only when the service answers "not found" or "protocol mismatch". The probes and scoring are identical for every protocol.

    Reasoning setting

    Messages keeps the original 8192 limit; Chat and Responses send no optional limit. Numbers beyond what the original prompt asks for are trimmed by the original rules. Hidden reasoning may also use tokens.

    Protocol or reasoning switches happen only when the service rejects a request before generating anything, so nothing is billed twice. If the model declines a probe (a short explanation and no numbers), automatic reasoning asks once more with the same settings; other requests that started generating, timed out or failed midway are never retried.

    2 model requests · Usually about 2,000 tokens each · Asked again once only if the model declines

    No installation or Cloudflare setup needed. Model calls use your API key. Results stay on this page unless you save them.

Loading the fingerprint database…
About the check

Start with two probes and add another if needed. For manual checks, send the prompts in order and paste the complete responses.

Check results

View the fingerprint comparison and suggested next steps.

Local analysis

Your conclusion will appear here

Connect an API or paste model answers to begin. The dial shows progress for each probe.

Results are for reference only and are not guaranteed to be accurate. They reflect behavioral similarity, and some closely related models may be indistinguishable.

FAQ

Common questions

Can I check whether ChatGPT or Codex has been downgraded?

It can help you investigate. Choose the manual check, send the probes to the ChatGPT or Codex you use, then paste the responses here. Results can suggest a model change, but a single check cannot establish that your account has been downgraded.

How can I check whether an API provider has switched models?

Enter your API URL and key, select the model, and start a check. The page compares the responses with the fingerprint database to assess whether they match the advertised model.

Does a mismatch mean the model has been replaced?

One mismatch is not enough to draw a conclusion. Follow the page's suggestion to run an additional probe and see whether the result is consistent.

Are two probes enough? When should I run another?

Start with two probes. If the comparisons disagree, the page will suggest another. If responses are incomplete, check that you pasted them in full. Similar models may still be indistinguishable.

How is the reference model selected?

Suggested from the selected model ID; you can change it. It helps compare the result without affecting scores or rankings.

Will the check use my API credits?

Yes. The default check sends two requests to your API, using an estimated 1,300–8,000 tokens at a reference cost of about $0.05. An extra probe adds one request and more usage. Actual charges depend on the model and your API provider.

Where are keys, answers and check records saved?

Use a temporary API key and revoke it afterward. In the default API mode, your key passes through the ModelDial relay; it is not written to databases, caches, logs or exported JSON, and its input is cleared when a check starts. Answers are analyzed in your browser. Save result stores a local record you can restore, delete or export.

Local record

Saved on request · Up to 5 records

Privacy & data

Your API key is used only for requests you initiate and is not written to a database, cache or logs. Probe copies, manual checks, JSON imports and API checks are counted separately, with model names, results and matching status. Relay checks also record token usage and anonymous channel identifiers. Original answers, free-form notes and API addresses are not stored. Details are retained for 90 days; daily aggregates are retained long term.

Responses are analyzed in your browser. They are stored here only when you choose Save result, and can be restored or deleted under Local records.

Complete security verification

Follow the instructions to verify and continue your request. Verification does not call a model.

Share