Bulk LLM Runner: Run ChatGPT, Claude, Gemini, DeepSeek and Qwen Prompts in Bulk (No API Key)

Run hundreds of prompts at once, or run ChatGPT-style AI on every row of an Excel or CSV file, across GPT, Claude, Gemini, Perplexity Sonar, the leading Chinese models (DeepSeek, Qwen, Kimi, GLM and MiniMax), Grok, Llama and 350+ other AI models. No OpenAI, Anthropic, Google or Perplexity API key needed. Built-in web search, PDF reading, image input, JSON output and side-by-side model comparison. Your results come back as a clean table you can download as Excel, CSV or JSON.

It replaces copy-pasting prompts into ChatGPT one by one, and pasting spreadsheet rows into an AI chat and copying the answers back by hand.

---

What you can do with it

• Run a list of prompts in bulk. Paste 10, 100 or 1,000 prompts. Up to 8 run in parallel, and every answer lands in its own row.
• Run AI on every row of a spreadsheet. Upload an Excel or CSV file, write one prompt with {{Column name}} placeholders, and get your file back with the AI answers as new columns. See how.
• Research with live web search. Any model can search the web before it answers, so you get current facts and source links, not training-data guesses.
• Get structured data, not paragraphs. Ask for "name, price and rating" and you get name, price and rating columns. For strict formats, supply a JSON schema.
• Compare ChatGPT vs Claude vs Gemini. Run the same prompts on several models side by side and pick the best one with real data.
• Read PDFs, web pages and images. Mention a URL or PDF in the prompt, or attach images for vision models.
• No API keys, no code. Your Apify account is the only login. Model usage is billed through Apify.

---

Run AI on every row of an Excel or CSV file

Have a spreadsheet with thousands of records and a question to answer for each one? Upload it, and the Actor runs your prompt once per row, with web search, and gives you your own spreadsheet back with the AI answers added as new columns.

How to set it up

1. Spreadsheet: upload your .xlsx, .csv or .tsv file, or paste a Google Sheets link shared as "Anyone with the link". The first row must hold the column names.
2. Prompts: write your question once, using column names in double curly braces. Each placeholder is replaced by that row's value: Laboratory: {{Lab Name}}, {{Address}}, phone {{Phone}}. Which of the checklist tests does this lab offer?
   • Column names are matched ignoring upper and lower case.
   • A prompt without placeholders still runs for every row, with the row's fields added below it.
   • Several prompts all run for every row (one output row per prompt).
   • The example prompts in the form are skipped automatically when a spreadsheet is attached.
3. Checklist (optional): one item per line, for example HbA1c, Lipid panel, Vitamin D 25-hydroxy. Each item becomes a column answered yes, no or unknown, with an evidence column and a source link column next to it, so you can spot-check any answer in one click. With a checklist you can leave Prompts empty: each row's fields are checked against the list.
4. Keep Enable web search on, pick a model and run.

Answer columns are named for you

Just ask your question in plain words. Before the run, the Actor works out the answer columns once, and every row uses exactly the same columns, so your file stays clean.

A file with First Name and Last Name and the prompt what year they were born, when they became president and what their net worth is gives you three new columns:

row_number	First Name	Last Name	Birth year	Presidential start year	Net worth
1	Donald	Trump	1946	2017	6.4B
2	Barack	Obama	1961	2009	Between $1.3 million and $10 million
3	Bill	Clinton	1946	1993	Between $5 million and $25.5 million

Want your own column names? Name them in the prompt and they are used exactly: Return birth_date, political_party and home_state gives birth_date, political_party and home_state columns. Full dates come back as YYYY-MM-DD in every row, and web-search citations are kept out of the cells.

What you get back

• Output tab: one row per spreadsheet row with row_number, all of your original columns, then the answer columns, then the prompt, model, response, cost and any error.
• Ready-made Excel file: open "Your spreadsheet with AI answers (Excel)" in the run's Output (stored as results.xlsx). It is your file in its original row order, with the answer columns added and the technical columns left out. This is the file to hand back to your team or client.

A checklist run looks like this:

row_number	Lab Name	Address	HbA1c	HbA1c (evidence)	HbA1c (source)	Lipid panel	...
1	Sonora Quest Laboratories	1255 W Washington St, Tempe, AZ	yes	Offers a diabetes panel including hemoglobin A1c.	https://www.sonoraquest.com/...	yes	...

Tips for accurate results

• The model answers "unknown" instead of guessing when it cannot find clear evidence. Those rows are the ones worth checking by hand.
• Test on 20 rows first. Compare the answers with a few records you already know, adjust the prompt or model, then run the full file. Search-capable models such as Gemini Flash and Perplexity Sonar give fewer unknowns than small general models.
• Rows run in parallel, so the Output tab lists them in the order they finished. Sort by row_number, or use the Excel file, which is already in your file's order.
• Stopped runs resume. Run again with the same input and finished rows are skipped.

---

Ready-made examples

Open one, click Try it, and the whole setup is filled in for you:

• Check which tests each lab offers from an Excel list: a spreadsheet of labs, a yes/no/unknown checklist, evidence and source links.
• Enrich a company list with founded year, HQ, CEO and size: company names in, research columns out.
• Write SEO product descriptions in bulk: one prompt per product, ready-to-paste copy.
• Classify and tag customer reviews at scale: sentiment, topic, urgency and a summary as separate columns.
• Compare GPT, Claude and Gemini answers: the same prompts on several models, side by side.

---

Quick start with a list of prompts

1. Paste your prompts into Prompts, one per line.
2. Pick a Model from the dropdown.
3. Optional: turn on Enable web search, add models to Compare with extra models, or attach Image URLs.
4. Click Start. When the run finishes, open the Output tab and download the results as Excel, CSV or JSON.

Input example

{
  "prompts": [
    "Write a 60-word product description for a waterproof hiking backpack. Return title, description and 5 keywords.",
    "Summarise the three biggest AI announcements of this week with a source link for each."
  ],
  "model": "anthropic/claude-haiku-4.5",
  "enable_web_search": true,
  "compare_models": ["openai/gpt-5-mini", "google/gemini-3.5-flash"]
}

Output example

One row per prompt and model. When a prompt asks for named fields, each field gets its own column:

{
  "prompt": "Write a 60-word product description for a waterproof hiking backpack. Return title, description and 5 keywords.",
  "model": "anthropic/claude-haiku-4.5",
  "title": "StormGuard 30L Waterproof Hiking Backpack",
  "description": "Built for wet trails and sudden downpours...",
  "keywords": ["waterproof backpack", "hiking daypack", "rain-proof rucksack", "30L backpack", "outdoor gear"],
  "tools_used": ["web_search"],
  "cost_usd": 0.0012,
  "error": null
}

A SUMMARY record in the run's key-value store totals the calls, tokens, failures and AI spend.

---

Features

Feature	What it does
Bulk prompts	Any number of prompts, up to 8 in parallel, one result row each
Spreadsheet mode	Excel, CSV, TSV or Google Sheets in, the same file with AI columns out
Checklist	Yes / no / unknown columns with evidence and a source link per answer
Web search	Live search results fed to any model before it answers
Auto JSON columns	Fields named in your prompt become separate spreadsheet columns
JSON schema mode	Forces the exact output shape you define
Model comparison	Every prompt runs on each extra model you add
Fallback models	If the main model fails or is rate-limited, up to 2 backup models take over
PDF and URL reading	The model reads PDFs and pages you mention in the prompt
Vision input	Attach images to every prompt for multimodal models
Reasoning effort	Turn reasoning up or down on models that support it
Automatic retries	Rate limits and temporary service errors are retried for you
Resumable runs	A stopped run continues where it left off
Per-call cost	Every row carries the AI cost of that call in cost_usd

---

Available AI models

The dropdown holds 350+ models and is refreshed regularly from the live model catalog, so new releases show up soon after launch. The main families:

Provider	Models
OpenAI (ChatGPT models)	GPT-5 family, GPT-4o, GPT-4o mini
Anthropic (Claude)	Claude Opus, Claude Sonnet, Claude Haiku
Google (Gemini)	Gemini Pro, Gemini Flash, Gemini Flash Lite
Perplexity	Sonar, Sonar Pro, Sonar Reasoning Pro, Sonar Deep Research
DeepSeek	DeepSeek V3 family, DeepSeek R1
Zhipu AI	GLM
MiniMax	MiniMax M series
xAI	Grok
Meta	Llama 4 Maverick, Llama 4 Scout, Llama 3.3 70B
Alibaba	Qwen
Moonshot AI	Kimi
Others	Mistral, Cohere Command, ByteDance Seed, Tencent Hunyuan and more

Need a model that is not in the dropdown? Type its id (in vendor/model form, for example qwen/qwen3.8-flash) into Custom model id.

Which model should I pick?

• Cheap, fast bulk work: a Flash, Flash Lite, mini or nano model, or DeepSeek and Qwen, which are strong for their price.
• Best writing and instruction following: Claude Sonnet or Claude Opus.
• Hard reasoning: the larger GPT-5 and Claude Opus models.
• Web research on every row: Gemini Flash or Perplexity Sonar, with web search on.
• Not sure? Run 10 prompts with Compare with extra models and let the results decide.

---

Who uses it

• Operations and research teams checking a fact for every business, lab, clinic or supplier in a spreadsheet.
• Sales and marketing teams enriching lead and company lists with founded year, headquarters, size or tech stack.
• Ecommerce and SEO teams writing product descriptions, meta titles, FAQs and ad variants for a whole catalog.
• Analysts classifying reviews, tickets, survey answers or emails into clean columns.
• Data teams turning unstructured pages and PDFs into structured JSON.
• Anyone choosing an AI model who wants to compare ChatGPT, Claude and Gemini on their own prompts.

---

Pricing

This Actor uses pay-per-event pricing: you pay for each result row the run produces, and the AI model usage of each call is billed through your Apify account. The current rates are shown on this Actor's Store page.

• No separate AI subscription or API key. You do not pay OpenAI, Anthropic, Google or Perplexity directly.
• Free Apify plan: runs are limited to 3 prompts (or 3 spreadsheet rows), model comparison is turned off, and AI usage is charged at 10 times the standard token rate. Subscribe to a paid Apify plan to run full lists at the standard rate.
• Spreadsheet mode makes one extra small call per prompt to name the answer columns. It does not create a result row.
• Failed calls are reported, not hidden. Each call is retried automatically. If it still fails, you get a row with the reason in the error column, and that row is billed like any other result.

---

Frequently asked questions

How do I run ChatGPT on every row of an Excel or CSV file?

Upload the file in Spreadsheet, write your prompt with {{Column name}} placeholders (or plain words, and the row is added for you), and run. The output keeps all of your columns and adds the AI answers next to them, and a ready-made results.xlsx gives you the file back in its original order.

Can I use ChatGPT, Claude or Gemini without an API key?

Yes. You don't sign up anywhere else. Your Apify account is the only login, and model usage is billed through it.

How many prompts can I run at once?

There is no fixed limit on the list. Up to 8 prompts run in parallel (set it with Parallel requests). Very large lists simply take longer. Free Apify plans are limited to 3 prompts per run.

Can I compare ChatGPT, Claude and Gemini on the same prompts?

Yes. Pick one model in Model, then add others in Compare with extra models. You get one row per prompt and model.

Does it search the web?

Yes. Turn on Enable web search and any model gets live search results before it answers. Perplexity Sonar models also search on their own.

Can the model read PDFs and web pages?

Yes. Turn on Enable PDF / file parsing and mention the URL in your prompt.

Can I attach images?

Yes. Paste direct image links into Image URLs and they are attached to every prompt, for models that accept images. Each link must open the image file itself, publicly, and be under 10 MB. Google Drive, Dropbox and OneDrive sharing pages show a web page instead of the image, so use the direct file link. Images are downloaded before the run starts, so a broken link stops the run right away with an explanation, and large images are resized automatically.

Can it return clean JSON or spreadsheet columns?

Yes. By default every answer is JSON, and fields you name in the prompt become columns. For a strict shape, set Response format to JSON Schema and paste your schema.

How do I see what each call cost?

Every row has a cost_usd column, and the SUMMARY record totals tokens and spend for the run.

Can I connect it to Google Sheets, Zapier, Make or n8n?

Yes. Use Apify's integrations or API to start runs and send results wherever you need them, and schedule runs to repeat automatically.

What happens to my data?

Your prompts go to the model provider you pick, to be answered. Most major providers state that they do not train on API traffic, but policies differ by provider, so check the provider's terms before sending sensitive data.

---

More AI Actors by the same author

• Bulk AI Image Generator: turn a list of prompts into images with Nano Banana (Gemini) and GPT Image.
• Bulk AI Translator: translate text lists, datasets, files or whole websites into any language.
• AI Brand Visibility Tracker: find out whether ChatGPT, Perplexity, Gemini, Claude and Grok recommend your brand.

💼 Need a custom workflow? Reach out at youssefarhan24@gmail.com