RSS

Recommender

Which AI model should I use? Describe your task

Describe in your own words what you want to do. The page looks for keywords and shows the models the data supports, with score, price and a cheaper or open option. No AI runs here: it is a keyword search.

Try:

Use cases with their own page

  • Summarize and write textSummaries, emails, reports and posts. The most common risk here is the model getting a name or a number wrong, so the chosen metric is factual accuracy.
  • Analyze documents and contractsReading contracts, PDFs and long documents, finding clauses and summarizing points of attention. Reasoning over dense text and a large context window matter here. It does not replace a lawyer's review.
  • Translate textTranslating websites, documents and messages. The catalog sources have no translation score, so the list shows the text models people actually use most, which is not a quality score.
  • Code and fix bugsWriting code, fixing bugs and reviewing pull requests. The most used yardstick is SWE-bench Verified, with real repository issues.
  • Agents and automationUsing tools, querying systems and finishing multi-step tasks without getting lost. Here the model acts, not just answers.
  • Customer support and chatbotsSupport chatbots that query systems and follow company policies. Since volume is usually high, the list only considers models with input up to US$ 1 per 1M tokens.
  • Reasoning and mathProblems that take several reasoning steps, such as math, logic and calculation. Models usually think a lot before answering here, which uses output tokens.
  • Answer factual questionsDirect knowledge questions with one right answer. What counts here is how often the model is right without searching the web.
  • Generate and edit imagesCreating illustrations, covers, banners and logos, or editing photos. The list follows the Arena text to image score, with the price from each company's official page alongside, plus an open weights alternative.
  • Generate videoShort videos from text or an image, such as reels, TikTok and ads. The list follows the Arena text to video score, with the price from each company's official page alongside, plus an open weights alternative.
  • Generate voice (text to speech)Narration, voiceover, dubbing and reading text aloud. There is no open voice quality leaderboard in the sources checked, so the list is ordered by price, not quality, with an open weights alternative.
  • Transcribe audioTurning meetings, interviews and videos into text, or making subtitles. There is no open transcription quality leaderboard in the sources checked, so the list is ordered by price, not quality, with an open weights alternative.

How the recommendation works

  1. The text is compared with a list of words and synonyms in Portuguese, English and Spanish, ignoring accents and case, accepting plurals and small typos in longer words.
  2. For text cases, the order follows the public metric available for the case, or real usage when there is none. For image and video, the order follows the Arena score, when there is one. For voice and transcription there is no common open leaderboard, so the order is by official price, and the page says so.
  3. The cost per task is an estimate you can adjust in the How we calculate block, with conversions and source shown. It does not change the order of the list.

Sources

  • Epoch AI, CC BY 4.0, latest version we checked.
  • OpenRouter Benchmarks, CC BY 4.0, latest version we checked.
  • OpenRouter, dados públicos da API; exports de benchmarks e rankings em CC BY 4.0, latest version we checked.
  • Hugging Face Hub, metadados públicos do Hub; licença de cada modelo no repositório, latest version we checked.

Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch. Family and version were grouped by rule from the name.