Hey, it's Cindy 🌱 You commented UNLOCK, so here are the three places from the reel, the direct links, exactly what each one is best at.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
Most people end up paying for four or five AI subscriptions because they never got to test the models properly before committing. These three fix that in different ways: one lets you talk to frontier models side by side in a browser, one runs open-source AI on your own computer with no cap at all, and one is a big lab's own playground with a genuinely usable free tier. Use them to find out what you actually need, then pay for that one thing.
at a glance
What each one is best for
🥊 Arena (formerly LM Arena) Best for comparing frontier models on your real task before you pay for any of them. Browser, nothing to install.
🖥 Pinokio Best for unlimited private generation: transcription, voice, images, video, running locally on your own machine with no per-use cost.
🔬 Google AI Studio Best for long-document work, image generation, and getting a free API key to wire into things you build.
💡 What none of them replace Your daily driver. These are for testing, private work and experiments. Client work and anything confidential stays in a paid account.
compare models
1. Arena, the model bake-off
Arena is the site the AI industry itself uses to rank models. You type one prompt and top models answer, then you vote on which answer was better. Those votes build the public leaderboard. It started as a university research project and the address is now arena.ai (the old lmarena.ai links redirect there).
🌐 Open arena.ai. You can start typing without an account. Signing in with a Google account raises your rate limits and keeps your history, which matters once you start running proper comparisons.
🎲 Battle mode is the default. You ask, two anonymous models answer, you pick the better one, and only then does it reveal which models they were. This is the mode that kills your brand bias, because you find out you preferred the answer from the model you were not planning to pay for.
⚖️ Switch to side by side to choose the models yourself. Use this when you have a shortlist and want a specific head to head rather than a random draw.
💬 Use direct chat for real work. One model, one conversation, no voting. This is the free way to spend an afternoon with a frontier model and see whether it fits how you think.
🏆 Read the leaderboard by category. Text, coding, image and video are separate boards for a reason. The best writing model is often not the best coding model, so check the board that matches the job you are hiring for.
The trade you are making. Arena's own notice says your conversations "will be disclosed to the relevant AI providers and may otherwise be disclosed publicly", and warns against submitting "personal information or other sensitive information that you would not want to be shared publicly". So test with a realistic task, never with real client data, unreleased work, or anything with a name and an address in it.
run it locally
2. Pinokio, your own AI computer
Pinokio is a free desktop app that works like an app store for open-source AI tools. You click install, it handles the environments and dependencies that normally make local AI a nightmare, and the tool runs on your machine. After the download there is no subscription and no usage cap, and your files never leave your computer.
⬇️ Download it. Get the installer from desktop.pinokio.co and pick your platform. Mac, Windows and Linux are all supported, and on Mac you choose Apple Silicon or Intel.
💾 Point it at your biggest drive. On first launch it asks where to keep its home folder. Model files are large, so give it real space, tens of gigabytes, on the drive that has room.
🔎 Filter the store by your hardware. This is the step people skip and then get frustrated. The store has filters for platform and for GPU (NVIDIA, AMD, Apple), so filter first and you only see apps that can run on your machine.
📦 Install one app and start it. Click download, then install, let the first big download finish, then hit start. The app opens as a local page in your browser, running entirely on your computer.
🔁 Come back for updates. The store is actively maintained and app versions update often, so check the launcher updates feed rather than reinstalling from scratch.
What is actually in there: the catalog is tagged, and the biggest tags right now are image generation (44 apps), text to speech (56), and video generation (38). A few concrete ones to look for: Uncensored Local Studio bundles local image generation, local language models, Whisper speech recognition and Kokoro speech synthesis in one offline app. Maestro is an all-in-one local video, image and music studio with a director mode that turns one prompt into a shot list (NVIDIA only, 6GB of video memory and up). AgentsView browses and searches your local Claude Code sessions, which is useful and needs no GPU at all.
Honest hardware talk: transcription and text to speech run on almost any modern laptop. Image generation wants a reasonable GPU or Apple Silicon. Video generation is the heavy one and will be slow or impossible on a light machine. If your laptop is a lightweight, the voice and transcription tools alone still replace a paid subscription or two.
Install hygiene. Stick to the apps listed in the Pinokio store rather than pasting random install links you found somewhere else, because these scripts run real commands on your computer. And if Windows flags an unsigned installer, read the warning properly before you click through instead of waving it away out of habit.
the lab's own playground
3. Google AI Studio, the underrated one
Google AI Studio is Google's own developer playground for its models, and you can use it in the browser with a Google account. It is not a polished consumer chat app, and that is exactly why it is useful: you get the raw model, a very large context window (a million tokens, so you can drop in whole books, transcripts or codebases), a prompt gallery of worked examples, and a free API key you can use in things you build.
🔐 Sign in and open a new prompt. Go to aistudio.google.com, sign in with a Google account, and start a new prompt. There is nothing to install and no card required to begin.
📚 Use it for the giant inputs. This is the thing it does that most chat apps will not let you do on a free plan: paste or upload something enormous (a full transcript, a long report, a year of notes) and ask questions across all of it at once.
🎨 Try the image model. Nano Banana 2, Google's current image model, is available in AI Studio and the API in preview according to Google's own announcement. It handles up to 14 reference objects and renders text in images properly, which is why it has become the default for product and mock-up work.
🔑 Grab an API key when you are ready to build. The free tier includes API access with rate limits, which is enough to test an automation end to end before you decide whether it earns a paid tier.
Read this before you paste anything sensitive. Google's terms draw a hard line between the free and paid tiers. On the free tier, "human reviewers may read, annotate, and process your API input and output", the content you submit is used to "provide, improve, and develop Google products and services", and the terms say plainly: "Do not submit sensitive, confidential, or personal information to the Unpaid Services." On the paid tier, Google states it "doesn't use your prompts or responses to improve our products" (the terms). Free is for your own experiments. Client work goes in a paid account.
the actual method
How to run a proper bake-off, then pay once
Free access is only worth something if it changes what you buy. Here is the version that takes an evening and saves you a year of stacked subscriptions.
📝 Pick your one real task. Not a riddle, not "write me a poem". The thing you would actually hand to a paid assistant every week: your weekly report, your ad copy, your client emails, your data cleanup.
🎯 Write one test brief and never change it. Use the prompt below. Same wording for every model, because the second you tweak it between tests, the comparison is worthless.
🥊 Run it in Arena, blind first. Battle mode, vote on the answer you genuinely prefer before you see the names. Do this three or four times with variations of your real task.
📊 Score the outputs, do not vibe them. Paste the outputs back into whichever model you trust most and use the scoring prompt below to force yourself into specifics.
💳 Buy one subscription, then use the free tools around it. One paid daily driver you use constantly, plus Pinokio for private and high volume jobs, plus AI Studio for the giant-input work. That stack costs one subscription, not five.
🧪 the test brief (paste the same one into every model)
I am testing you against other AI models on a task I do every week, so treat this as a real work request and not a demo. The task: [describe the exact task, for example "turn my messy meeting notes into a client-ready weekly status update"]. Here is the real input: [paste your actual notes, brief, data or draft].
Produce the finished piece of work, in the format I would actually send: [name the format, for example "a 200 word email with a one-line summary, then wins, blockers and next steps"]. Match this tone: [describe your tone in a few words, or paste two sentences you have written before].
Then, underneath the finished work, give me three short notes: what you assumed because I did not tell you, what you would need from me to do this better next time, and the one part of this output you are least confident about. Do not pad the answer with preamble or a summary of what you are about to do.
📊 score the outputs against each other
I ran the same brief through several AI models and I am deciding which one to pay for. Below are the outputs, labelled A, B, C. Score each one out of 5 on these five criteria and show your reasoning in one line per criterion: accuracy (did it invent anything that is not in my input), format compliance (did it produce exactly what I asked for), tone match, usefulness of the assumptions and questions it raised, and how much editing I would need to do before sending it.
Then give me a table of the totals, name the winner, and be specific about what the winner did better rather than saying it was more comprehensive. Finally, tell me which one you would pick if my priority was [speed / accuracy / writing quality / cost] instead, since that changes the answer.
Outputs:
A: [paste output A]
B: [paste output B]
C: [paste output C]
the honest bit
Why these are free
Nothing here is charity, and knowing the mechanism tells you where the edges are. Arena is funded research and your prompts and votes are the data that builds the leaderboard, so treat everything you type there as potentially public. Pinokio is genuinely free because the compute is yours: your electricity, your GPU, your hard drive, no one else in the loop. Google AI Studio's free tier is a developer funnel, and the price of admission is that your unpaid-tier content can be reviewed by humans and used to improve their products.
None of that makes them less worth using. It makes them worth using for the right things: comparing, experimenting, and doing volume work that does not touch anyone else's data. Then you pay for one tool, properly, and use it every day.
Claude (my paid daily driver, and the free plan is a real one): claude.ai
Want to stop collecting tools and start building with them? Inside AI System Club I teach people to learn AI from scratch and build real systems, for work and for their own business. Doors open soon and founding pricing dies at close. 🌱 Join the waitlist (link in my bio too).
the waitlist is now open
AI System Club is launching soon
Build AI systems with me through recorded lessons you can follow at your own pace. Join the waitlist for first access and the founding price.
btw this list gets first dibs the moment doors open :)