The model you already use is probably not the best one for this job.

WhichAI.today exists so anyone can match a task to the AI model that is actually leading that kind of work today, instead of defaulting to whichever chatbot they signed up for first.

Our mission

Most people treat "AI" as one product. It is not. Coding, contract review, creative writing, math, high-volume automation, and everyday questions are different jobs, and they are won by different models. The mission is to make that choice obvious in one screen: the current leader, why it leads, a strong alternative, and the best model you can run yourself.

AI leadership changes constantly

Frontier models leapfrog each other every few weeks, and they do not leapfrog in the same category. A model that leads coding this month can trail on legal reading or speed. Last quarter's default is often this month's second or third choice. These are the leaders as of October 24, 2026. The table updates when the rankings do.

Why choosing in real time matters

• Quality. The gap between the leader and the fifth-place model on a hard coding or legal benchmark is large enough to change whether the work is usable.
• Cost. The same task can cost 10 to 100 times more on a flagship model than on a fast model that is good enough. Volume work should not run on the most expensive brain.
• Speed. A support bot and a math proof do not want the same latency. The fastest model is rarely the smartest, and the smartest is rarely the fastest.
• Privacy. Client files, source code, and contracts sometimes cannot leave your network. That is a different winner: the best open-weights model you can host yourself.
• Habit. The expensive mistake is staying on last year's default because it is already open in a tab. The rankings exist so that habit has a cheap way to be checked.

How WhichAI.today works

Every recommendation comes from one of six categories: coding, legal and long documents, writing and marketing, reasoning and math, speed and high volume, and a general catch-all for everyday tasks. Each has its own page with the current picks and the leader history. Picks are grounded in public evaluations from LMArena, Artificial Analysis, Vals AI, SWE-bench, GPQA, and Humanity's Last Exam.

A daily job re-checks those sources. A new number one has to show up on two daily checks before it replaces the current leader. When you search, common phrases match a category instantly. Unusual wording goes to a fast classifier, and anything it cannot place lands in General, so a search always returns a recommendation. Plan a project does the same thing for a whole piece of work: it splits the description into stages and labels each stage with a category. The model names always come from the rankings, never from the planner's own opinion.

How to use it

Quick pick is for one job. Try "fix a bug in my React app," "summarize this earnings call," or "write a LinkedIn post." You get today's leader, an alternative, and an open-weights option, each with a reason and a direct link.

Plan a project is for work with more than one kind of task. Describe what you are building, the main parts, the timeline, the team, and the scale. A solo blog relaunch, a three-month invoicing app, and a 40-person contract-review rollout each come back as a different sequence of stages, with the right model for that stage.

Who runs this

Run by Pierce D.. Report an error on GitHub.

Independence

No paid placement, no affiliate links in the picks, and no provider sponsorship. A model is listed because of the public board for that job.

Corrections

Open a GitHub issue to report an error. Corrections and changes to the ranking method are logged publicly on Changes.

Methodology and limits

The boards, the two-check rule, and the fact-check gate are on the methodology page. Benchmarks measure a public task, not your exact files. Rankings can lag a launch by a day, because the refresh runs once a day and only publishes answers that pass the checks.

Questions people ask