NinA Benchmarks
AI models compared on the work businesses actually do.
we keep adding new ones
Most benchmarks are written for researchers: cryptic names, scores without context. We test models on real-world tasks and show the outcome in a way you can use. Which model fits which work, and what does it cost?
01Benchmark · updated 9 November 2026
Which AI model offers the best value for money?
How capable a model is, set against what one task costs on average.
How to read it
- The higher the dot, the better the model performs.
- The further left, the lower the average cost per task.
- The model with the best value for money is highlighted.
- Drag a box across the chart to zoom in. Double-click or the "Herstel" (reset) button resets it.
What it means for you
The most expensive model is not always the best. For tasks you run thousands of times a month, a model that is almost as good at a fraction of the price saves more than the quality gap is worth.
02Benchmark · updated 9 November 2026
Which AI model is best for automating your business workflow?
How well a model automates a business workflow.
How to read it
- Each bar is the average score of one model.
- The score is the share of task objectives completed with no guardrail violations.
- The benchmark covers 657 tasks across six business domains (Finance, HR, Marketing, Operations, Sales, Support) using simulated environments for applications including Gmail, Google Sheets, Slack, Salesforce, Zendesk, Jira, and HubSpot.
- Higher is better.
- Point at a bar for the details.
What it means for you
A model at the top of a general ranking is not automatically at the top here. Which model works best depends on the work you give it. That is why we choose per use case.
03Benchmark · updated 9 November 2026
Which AI model is best for business assignments?
Models compared head to head on business tasks, expressed as an ELO score.
How to read it
- Each bar is one model. The test consists of 91 tasks from four projects lasting several weeks, such as data science, product management, banking operations and strategy for heavy industry.
- For each task the model gets many source files and has to deliver an end product, such as a spreadsheet, a presentation or a memo. That is real knowledge work, not a quiz question.
- The result is judged in three ways: with a checklist (is the right information used and are the conclusions correct) and with two comparisons against another model's work: who works more thoroughly and sharply, and who delivers the more professional result.
- The ELO score combines those three. ELO works like a chess rating: whoever wins more often rises. Higher is better.
- Point at a bar for the upper and lower bound. If two models overlap, they are practically equally good.
What it means for you
Small score differences say little. Look at the margins: when two models are close, price, speed and where your data is stored decide.
04Benchmark · updated 9 November 2026
Which AI model gets figures out of your spreadsheets and Excel documents best?
How reliably a model processes figures from spreadsheets and documents, set against what one task costs on average.
How to read it
- Each dot is one model. The test consists of 80 questions like a data analyst gets every day: pulling numbers out of Excel files and giving an exact answer.
- The higher the dot, the more reliable the model. Every question was asked five times and only counts if all five answers are correct.
- The further left, the lower the average cost per task. That is what the model costs in total to answer one question, including all the steps along the way.
- The model with the best value for money is highlighted.
- Drag a box across the chart to zoom in. Double-click or the "Herstel" (reset) button resets it.
What it means for you
For work with numbers, such as reporting, controlling and invoices, what matters most is that a model gets it right every time, not now and then. It remains a test: always try it with your own files first.
05Benchmark · updated 9 November 2026
Which AI model is the most reliable and makes things up the least?
How much a model really knows, and how often it guesses when it does not know something.
How to read it
- Each bar is the score of one model. The test consists of 3,000 questions on business, law, health and software, among other things, which the model has to answer from memory, without files or looking things up.
- Higher is better. The score is the share of correct answers minus the share of wrong answers. Saying you do not know does not count as wrong, so guessing costs points.
- Point at a bar for the details: how many answers are correct (higher is better) and how often the model makes something up when it does not know (lower is better).
What it means for you
A model that knows a lot is not yet a reliable assistant. If it guesses when in doubt, you get confident but wrong answers, and those hurt in customer questions and legal or financial texts. So do not only look at who has the most correct answers, but also at who most often dares to say it does not know. Which model fits depends on what a wrong answer costs you.
06Benchmark · updated 9 November 2026
Which AI model completes a work task best?
How well an AI agent completes an entire work task, in real software.
How to read it
- Each bar is one model. Under the name is the software that runs the model (for example Codex or Claude Code). That is called a harness: the agent's working environment.
- The test consists of 122 tasks from the work of domain experts: multiple steps, in real software, with a result that a computer checks.
- There are two figures. The pass rate only counts a task if everything is correct. The score also counts how much is correct. A report with ten parts of which nine are correct counts as failed in the pass rate and as 90% in the score. Higher is better.
- Switch between the two at the top right, or choose "Beide" (both): the hatched bar is the score, the solid bar the pass rate.
- Point at a bar for the details, including the cost per task.
What it means for you
Even the best agents do not fully succeed at every task. A process that is 90% right still has to be checked. So look mainly at the pass rate and build in control steps. The chosen model makes more difference than the software around it, so compare models for your own process.
07Benchmark · updated 9 November 2026
What is the best AI model for reading (long) PDF documents?
How often a model gets everything right on long documents, set against what one task costs on average.
How to read it
- Each dot is one model. The test consists of 100 tasks across ten fields. The model gets a long PDF document, such as a contract or an insurance claim, and an assignment. It is then judged on a list of concrete points.
- The higher the dot, the more often the model gets all points right. If it misses one, the attempt counts as failed. Every task was run five times.
- The further left, the lower the average cost per task.
- The model with the best value for money is highlighted.
- Drag a box across the chart to zoom in. Double-click or the "Herstel" (reset) button resets it.
What it means for you
A lot of client work consists of reading documents: contracts, policies, manuals and claims. Even the best model gets everything right in only about a third of attempts, so expect the result to need checking, even if it looks good. Also look at the price: models that score almost as well sometimes cost much less per task.
Next
A benchmark does not choose a model for you. We do.
Which model is best depends on your work, your data and what it may cost per month. In an AI Partnership we choose per use case, and switch when a better model comes out.
More on how we approach that: the AI Partnership.
Free to use: all benchmarks are licensed under CC BY 4.0. Reuse or quote them, crediting: NinA AI Benchmarks, nina-ai.nl/benchmarks.
Which model fits you?
Let us compare them on your work.
A 15-minute intro call. Afterwards you know which models and which approach fit your processes.
- 160+ organizations
- Sessions rated 9.3
- Amsterdam







