Prompt every LM Studio model and log time and readability

Prompt every LM Studio model and log time and readability

An n8n workflow you drive from a chat box. Each message you send goes to every model that your LM Studio server lists.
The workflow times each reply, scores how easy it is to read, and adds one row per model to a Google Sheet. You end up
with a table that compares the models on the same prompt.

What is inside

• workflow.json: the workflow as n8n exports it, 12 nodes and 8 sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The server address was a private one. Get Models and Run model with dynamic input now read
  http://<YOUR_LM_STUDIO_HOST>:1234/v1.
• Two node names had typos. Extract Model IDsto Run Separately is now Extract model IDs, and
  Run Model with Dunamic Inputs is now Run model with dynamic input. Every expression that named them follows.
• Get timeDifference asked n8n for the gap between the two times without a unit, and n8n's source gives days as the
  default, so each reply would have logged as 0. It now asks for seconds, and Prepare Data for Analysis stores that
  number.
• The Google Sheets step mapped columns whose names matched no header list, and it pointed at the author's own
  spreadsheet. It now maps eleven columns by name, and the document is <YOUR_SPREADSHEET_ID>. The sheet's header row is
  listed in a sticky note.
• The workflow's time zone and call policy settings were the author's, and they are gone. Times follow your n8n instance.
• Two sticky notes held the author's private address and a second copy of the setup list. One is gone, and one is
  rewritten to say where the address goes.

Before you start

• An n8n instance that can reach the machine that runs LM Studio, on n8n Cloud or your own server.
• LM Studio with its local server running and the models you want to compare available.
• A Google account for the sheet. The sheet step is optional.

Set it up

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. In Get Models and Run model with dynamic input, replace <YOUR_LM_STUDIO_HOST> with the address of the machine that
   runs the LM Studio server. The port in the file is 1234.
3. Add an OpenAI credential and select it in Run model with dynamic input. The node needs some key value. Any text
   works when your LM Studio server asks for no token.
4. Create a Google Sheet and put these eleven headers in its first row, spelled as shown:
   - Prompt: the message you sent
   - Time Sent: when the request went out
   - Time Received: when the reply came back
   - Total Time Seconds: the seconds between the two
   - Model: the model's id
   - Response: the reply's text
   - Readability Score: the Flesch reading ease estimate
   - Average Word Length: letters per word
   - Word Count: words in the reply
   - Sentence Count: sentences in the reply
   - Average Sentence Length: words per sentence
5. In Save Results to Google Sheets, pick your Google Sheets credential, replace <YOUR_SPREADSHEET_ID> with your
   spreadsheet's ID and check that the sheet is called Sheet1. To skip the sheet, delete that node.
6. Open the chat, type a prompt and send it.

How a run goes

1. When chat message received takes your message as chatInput.
2. Get Models asks the server's /v1/models endpoint for its list.
3. Extract model IDs splits the list into one item per model.
4. For each model, Capture Start Time notes the time and Add System Prompt adds the system prompt, which reads
   Ensure that messages are concise and to the point readable by a 5th grader.
5. LLM Response Analysis sends your message and that system prompt to the model. Run model with dynamic input is the
   chat model behind it, and it takes the model id from the item. Its options are temperature 1, top P 1, presence
   penalty 0 and a timeout of 250 seconds.
6. Capture End Time notes the time after the reply, and Get timeDifference works out the seconds between the two.
7. Prepare Data for Analysis gathers the prompt, model, reply, both times and the seconds.
8. Analyze LLM Response Metrics counts words and sentences, averages the word and sentence lengths and works out a
   readability score. The score is the Flesch reading ease formula, with syllables estimated from groups of vowels.
9. Save Results to Google Sheets appends one row per model.

Change the look

• The system prompt: the text in Add System Prompt.
• The settings of every model: Run model with dynamic input, for temperature, top P and presence penalty.
• Which models run: add a Filter node after Extract model IDs to drop the ones you do not want to test.
• What is measured: the code in Analyze LLM Response Metrics.
• Where the results go: replace Save Results to Google Sheets with another storage node, or delete it and read the
  results in the run.

Good to know

• The facts below come from pages and n8n source read on 12 November 2026.
• LM Studio's page on its models endpoint says the list may include all downloaded models when Just-In-Time loading is
  on. The workflow then prompts every one of them, so the run can take a long time. Use the Filter node above to narrow it.
• LM Studio's page on authentication says it asks for no token by default. If you turn the requirement on, `Run model with
  dynamic input sends your credential's key, and you must add a Bearer header to Get Models` yourself.
• The time runs from just before the request to just after the reply. Anything LM Studio does in between, such as loading
  a model, is inside the number.
• The readability score is an estimate. Its syllable count looks at vowel groups, not at a dictionary. It suits comparing
  models on one prompt and is a poor guide to any single text.
• There is no memory in the chat, so each message starts fresh.
• A run goes through the models one after another, and a model that takes longer than 250 seconds stops the run with an error.
• Billing: the models run on your own machine, so no model provider bills per call.
• Your data: your prompt and every reply go to the LM Studio server you name and, in the sheet step, to Google.
• Credentials: the JSON holds none.