The AI ecosystem grows every week. It's easy to end up subscribed to three services that do the same thing, or testing 40 wrappers that shut down in six months. This guide is not a "top 200" list; it's a method to filter for yourself.

Four criteria to evaluate any AI tool

  1. Does it use a top-tier base model? Useful AI tools usually rely on Claude (Anthropic), GPT (OpenAI), Gemini (Google), or a serious open-source model like Llama or Mistral. If the tool hides which model it uses, be suspicious.
  2. Does it add value on top of the API? If the answer is "it's a textbox that calls GPT-4", it's probably cheaper to call the API directly. Good vertical AIs (legal, medical, code, audio…) add workflows, proprietary data, or ergonomics the API doesn't provide.
  3. Does it have a credible business model? "Free forever" usually means "this shuts down in six months" or "you are the product". A reasonable monthly subscription is usually sustainable.
  4. Does it export your data? Any serious tool lets you export prompts, generated files, and results. If you can't export, you can't leave.

Categories and references by area

Text and reasoning

  • Claude (claude.ai) — long-form reasoning and writing. Excellent for prose refactoring.
  • ChatGPT — the Swiss-army knife. Plugins, Code Interpreter, image generation included.
  • Perplexity — search with citations. Replaces Google for source-backed questions.

Image

  • Midjourney — the reference for concept art.
  • Leonardo AI — alternative with a decent free plan and fine style control.
  • Stable Diffusion — open source, self-hostable. For full-control users.

Audio

  • ElevenLabs — synthetic voices nearly indistinguishable from humans. Market leader.
  • Whisper (OpenAI) — open-source transcription. Works via CLI or Python.
  • Suno and Udio — full music generation with vocals.

Video

  • Runway — VFX, rotoscoping, and video generation.
  • Descript — edit video by editing the transcript text.

Code

Three frequent traps

1. Subscribing to an expensive wrapper

If a tool charges 30 €/month for features the base Claude or OpenAI API offers for 5 €, it's probably a wrapper with little personality. Before paying, ask yourself what it does that the API doesn't.

2. Mistaking hype for usefulness

A tool viral on Twitter isn't necessarily a useful tool. The honest metric: will you still be using it three months from now? If the answer is "not sure", don't subscribe yet.

3. Ignoring energy cost and privacy

Every call to a large model consumes electricity, and if you work with sensitive data (clients, IP), you must read the provider's terms about training on your data. Serious services offer opt-out or zero-retention plans.

Recommended pattern for using AI in a project

  1. Start with the base model directly (no wrappers) until you understand what you really need.
  2. If a recurring need arises, evaluate specialised vertical tools.
  3. Combine two or three quality tools rather than ten generic ones.
  4. Always keep prompts in version-controlled files — they're as much code as the rest.

For commercial workflows (digital products, marketing content, visual branding), the VBedits store includes tools that combine several of these AIs into opinionated flows, so you don't reinvent the wrapper yourself.