WoAI Rating
All models →Overall rating across nine build categories · Higher is better
Independent AI model ratings · 86 models rated
Which AI model should you build with?
One rating for every new model, based on real build work: web apps, games, 3D, SVG and agent tasks. Then prove it on your own prompts with your own API keys.
Free to start, no card. Made by the @intheworldofai YouTube channel.
NEW · AI NEWSKnow which models are coming before they ship
Model leaks, new models spotted before launch and lab news in one feed, updated every 15 minutes.
NEW · LEADERBOARDClaude Opus 5.5 leads the board at 80.0Ratings now sit on a hard scale, so the gap between frontier and open models is real.
NEW · BUILD RACERace up to four models on the same appEach one builds, gets checked in a real browser, and fixes its own bugs.
Rating the newest models from
Frontend & UI
Category board →Web apps, landing pages and dashboards · Higher is better
Agentic
Category board →Multi-step tool use and long tasks · Higher is better
Benchmark your own prompts
Every output is checked in a real browser and scored by a panel of judges from different labs. You pay providers directly; we never mark up inference.
Build Race
Pick up to four models, describe the app in plain English, and watch them build, get checked and fix their own bugs.
Changelog
Docs →- Oct 11 · FeatureAI News: leaks and new models before they launch
- Oct 11 · RatingsHard scale: the best model now rates about 80, not 95
- Oct 11 · ScoringRuns scored by a judge panel plus real browser checks
- Oct 11 · FeatureMy Leaderboard: your models, ranked by your prompts
86models rated, from frontier to open weights
120real build tasks behind every rating
9scored categories, from frontend to agents
0%markup on your API usage, ever
AI News · Pro
Know what’s coming before it ships.
Model leaks, new models spotted before launch and lab announcements land in one feed, so you can test the next model the day it drops.
- LeaksPosts from the accounts that find new models first.
- New models, earlyFresh listings on OpenRouter and Hugging Face, often before the launch post. Benchmark them in one click.
- Lab newsOfficial announcements from the big labs, plus the AI stories people are talking about.
- From World of AIOur own write-ups and the latest videos from the channel.
Get AI News with Pro Updated every 15 minutes
AI NewsChecked 3m ago
- LeakNew model id spotted in a public API changelog, with a 1M context windowPosted on X · 12m ago
- New modelFresh open-weights coding model appears on Hugging FaceSpotted before launch · 40m ago
- Lab newsA frontier lab announces its next flagshipOfficial announcement · 2h ago
- World of AIEverything we know so far, and what to test firstPost · 3h ago
Sample of the feed
Pricing
Browse free. Go Pro when you’re choosing for real.
You bring your own API keys, so there are no credits or usage tiers. Pro unlocks everything; Pro+ adds instant leak alerts and longer history.
Free
$0
No card needed
- Public leaderboard and model pages
- One benchmark preview with the generated output
- Can I Run It hardware tool
- Docs and community
Pro
Most popular$12/ month
Cancel any time
- Full leaderboard with every category breakdown
- AI News: model leaks, new models before launch and lab news, every 15 minutes
- Unlimited benchmarks on your own API keys
- Browser checks, judge panel and a full breakdown for every score
- My Leaderboard, run history and CSV exports
- Build Race, Playground, Model Advisor and Plan My App
Pro+
New$29/ month
Cancel any time
- Everything in Pro
- Instant leak alerts: an email the moment a model leak or World of AI post lands
- A full year of benchmark history (Pro keeps 30 days)
- Model Watch on up to 10 prompts with a $100 monthly cap
The leaderboard
The WoAI Rating
Curated by World of AI from public evaluations and hands-on testing. It is hard on purpose: no model is near 100 yet, so real gaps show up.
Top 20 · Overall
Scale 0–100 · each overall rating is the task-weighted average of nine categoriesHow ratings work →
| # | Model | Rating | Frontend | Agentic | ||||
|---|---|---|---|---|---|---|---|---|
Unlock all 86 models and every category score with Pro.Start freeUnlock with Pro | ||||||||
Head to head
See where each model wins
Pick any two of the top 20. Bars show the WoAI Rating overall and in five build categories; the winner of each row takes its lab’s color.
vs
Pro compares all 86 models side by side, with cost, speed and reasoning levels.Compare more models
Inside the app
Everything you need to pick a model, in one place
The rating, your own leaderboard, live runs and Build Race sit on one dashboard, in light or dark.
Rating at a glance
Filter the board by category, open weights or best value, and click any bar for the model page.
Your own leaderboard
Every run you score lands on My Leaderboard, so your models are ranked by your work.
Live runs and races
Watch benchmarks and Build Races progress from the top bar while you keep working.
Methodology
How the rating works
One number you can compare at a glance, with the working shown on every model page.
01
Real build tasks
120 tasks across nine categories: frontend, games, 3D, SVG, systems, agents, reasoning, debugging and creative work.
02
Evidence, then testing
Public evaluations from across the field, checked against hands-on runs. Every model page shows how often WoAI ran it.
03
One hard number
The nine category ratings are averaged by task count. The scale is strict, so 80 is the frontier and a weak model scores far lower.
Your benchmarks
The board tells you where to start. Your prompts tell you what to ship.
Run any model you have a key for against your own prompts or the WoAI suite. Every score shows its working.
- Your keysYou pay providers directly. WoAI never resells inference or marks it up.
- Real browser checksEach output runs in a locked-down browser: does it load, throw errors, fit a phone, animate, respond to input.
- A panel of judgesJudges from different labs score five dimensions. The lab that made the model sits out whenever you have another key.
- Your own leaderboardEvery run lands on My Leaderboard, so your models are ranked by your work, not ours.
Example resultGame dev · Pong clone
78final score
= 30% browser checks (83) + 70% judges (76)
Browser checks · 5 of 6 passed
- ✓Renders visible content
- ✓No runtime errors
- ✕No console errors
- ✓Fits a phone screen
- ✓Animates on its own
- ✓Handles clicks and keys
Judge panel · 96% agreement
- Gemini 3 Flash74
- GPT-4.178
FAQ
Questions, answered
Is it really free to start?
Yes. Create an account without a card to see the top of the leaderboard, open model pages and try one benchmark preview. Pro unlocks every model, every category score and unlimited runs on your own keys.
Do I pay for the AI usage?
You use your own API keys, so providers bill you directly at their normal prices. WoAI never resells inference or adds a markup, and your keys are never shown back to the browser.
Why is no model near 100?
The rating is hard on purpose. A strict scale keeps real gaps visible: today’s best model rates about 80, strong models sit between 55 and 70, and weaker models fall well below that.
How is the WoAI Rating made?
It is curated by World of AI from public evaluations and hands-on testing across 120 build tasks in nine categories. Each overall rating is the task-weighted average of the nine category ratings.
What does Pro add?
The full leaderboard with every model and every category score, unlimited benchmarks on your own keys with browser checks and a judge panel, My Leaderboard, run history, CSV exports, Build Race, AI News (leaks, new models and lab news, updated every 15 minutes) and the planning tools.
What is AI News?
A Pro feed of model leaks from the accounts that find them first, new models spotted on OpenRouter and Hugging Face before launch, official lab announcements, and posts and videos from World of AI. It updates every 15 minutes, and a new model links straight to a benchmark when it is available.
Can I cancel any time?
Yes. Manage or cancel your plan from the billing page in your account. You keep Pro until the end of the period you paid for.
Pick your next model with evidence, not vibes.
Free to start. Bring your keys when you’re ready to test.