Benchmarks don't know your task
A model that tops a leaderboard can still be the wrong choice for your prompts. Testing on your own question, your own data and your own tone is the quickest way to see which answer you would actually ship.
ModelCompare: test AI models side by side on your own prompt
ModelCompare is a free-to-try web app for comparing large language models. Type one prompt, send it to up to 6 models at once, and read their answers next to each other, with how fast each one replied and what each answer cost. It covers 30 models from 14 labs, including OpenAI, Anthropic, Google, xAI, DeepSeek, and Mistral AI, with no API keys, accounts or subscriptions needed to start.
What ModelCompare does
Choosing an AI model usually means reading leaderboards, opening several chat apps and pasting the same question into each one. ModelCompare puts that in one window: pick the models, write the prompt once, and every model answers in its own column at the same time.
Next to the answers you get the numbers that matter when you pay per token: time to first token, total time, output speed, tokens used and the exact cost of each answer. Follow-up messages go to every model, so you can compare whole conversations, not just one reply.
To narrow the list before you test, a price vs. intelligence map plots every scored model by its Artificial Analysis Intelligence Index and its blended price per million tokens, so the best value models stand out.
Why it helps
See the real cost before you commit
Prices per million tokens are hard to picture. ModelCompare shows an estimate before you send and the exact cost of every answer after, so you can tell whether a model that is 20 times cheaper is good enough.
Speed is part of quality
For chat and agents, how long you wait matters. Each run records time to first token and tokens per second, measured on the same prompt at the same moment.
One place instead of a dozen accounts
No API keys, no subscription per lab and no setup. You pay only the model's own per-token list price, from credits that start free.
Features
Side-by-side answers
Send one prompt to up to 6 models and read the answers in parallel columns, streamed live.
Speed and cost per answer
A summary table lists time to first token, total time, tokens out, speed and cost for each model.
Price vs. intelligence map
Every scored model on one chart, split into the sweet spot, the frontier, workhorses and the hard to justify. Switch “Rank by” to coding, science, hard reasoning, tool use or long documents to re-rank the map by that benchmark.
Search and filters
Find models by name, lab or feature, and filter by open weights, vision, reasoning, price under $1 or 1M+ context.
Images and PDFs
Attach one image or PDF to a prompt to compare how models read charts, screenshots and documents.
Reasoning effort
Set the reasoning effort once for every model in the comparison that supports it.
Shareable links
The address of a comparison names its models, so you can send the matchup to a teammate. In chat apps and social posts the link previews as the models on the price vs. intelligence map, with their scores and prices.
History
Past comparisons are saved to your history, where you can search and reopen them.
How-to guides
Short walkthroughs for the most common jobs.
How to compare AI models side by side
- Open ModelCompare. Two models are already picked so you can start right away.
- Add or swap models in the models panel. Search by name or lab, or load a quick set such as Smartest 3 or Cheapest 3. You can compare up to 6 at once.
- Type your prompt, or click a starter prompt for coding, writing, data, reasoning or summarising.
- Press Send. Every model answers in its own column at the same time.
- Read the summary table for speed and cost, then send a follow-up to keep comparing.
How to find the best AI model for your budget
- Open the price vs. intelligence map.
- Look top-left at the sweet spot: models that score above the median and cost under $2 per million tokens.
- Click a model to open it in a comparison against others.
- Run your real prompt and check the cost column. If a cheaper model's answer is good enough, you have your pick.
How to test models on an image or PDF
- Turn on the Vision filter in the models panel to keep only models that read images.
- Click Attach, or drop a file onto the prompt box. One image or PDF per message.
- Ask your question about the file and press Send.
How to change reasoning effort
- Open the reasoning effort menu next to the prompt box.
- Pick a level. The menu tells you how many of the models in the comparison support it.
- Send the prompt and compare quality, speed and cost at that effort.
How to share a comparison
- Pick the models you want to show.
- Click Copy link under the heading, or copy the address from your browser.
- Send the link. Whoever opens it gets the same models, ready for their own prompt. The map starts folded into a bar of those models, and Change models opens it again.
How to get more credits
- Everyone starts with $0.05 of free credits, no sign-in needed.
- Sign in with Google to raise your free credits to $0.20.
- When those run out, open your usage and buy more credits. Each answer only uses the model's per-token list price.
Frequently asked questions
What is ModelCompare?
ModelCompare is a web app that sends one prompt to several AI models at once and shows their answers side by side, with the speed and cost of each answer. It covers 30 models from 14 labs, including OpenAI, Anthropic, Google, xAI, DeepSeek, and Mistral AI.
Is ModelCompare free?
You can start for free. Everyone gets $0.05 of credits without an account, and signing in with Google raises that to $0.20. After that you buy credits and pay only the model's per-token list price for each answer. There is no subscription.
How much does one comparison cost?
It depends on the models and how long the answers are. Each answer is charged at that model's list price per input and output token. The prompt box shows an estimate before you send, and the summary table shows the exact cost of every answer after.
Do I need API keys or accounts with OpenAI, Anthropic or Google?
No. ModelCompare runs the models for you, so you don't need an API key or a subscription with any lab.
How many models can I compare at once?
Up to 6 models in one comparison. Every follow-up message goes to all of them.
Which AI models can I compare?
30 models from 14 labs, including OpenAI, Anthropic, Google, xAI, DeepSeek, and Mistral AI. The catalog keeps the newest model of each family and is refreshed as labs release new models and change prices.
Can I compare how models read images and PDFs?
Yes. Attach one image or PDF to a message. Models that can't read images are flagged, and the Vision filter shows only the ones that can.
Where do the intelligence scores come from?
From the Artificial Analysis Intelligence Index, which combines results from several reasoning, knowledge, maths and coding benchmarks into one score. Newly released models may not have a score yet. On the map you can also rank by a single area, such as coding, science, hard reasoning, tool use or long documents, using Artificial Analysis's per-benchmark scores.
How is the price on the map calculated?
It is a blended price per million tokens: three parts input to one part output, which is close to a typical chat. The exact input and output prices are listed for every model.
Can I share a comparison?
Yes. The page address lists the models being compared. Click Copy link and send it, and the other person sees the same matchup.
Which languages does ModelCompare support?
The interface is available in English, Arabic, Spanish, Chinese, Hindi, French and Russian. You can prompt the models in any language they understand.
Try it on your own prompt
The fastest way to pick a model is to watch a few of them answer the same question.