Jahanzaib

AI Models

900 models from 72 labs, with the price at every provider that serves them, refreshed daily, plus independent scores on the 70 that matter most. Updated Oct 15, 2026.

CompareRankings

900 models

Browse by lab

Each lab page lists its models with prices, context and release dates on one screen.

How this directory works

  • Catalog data (prices, context, modalities, capabilities, release dates and the hosts that serve each model) comes from models.dev, an open database released under the MIT license, and is pulled again every day. The headline price is the lab's own; every other host is listed on the model page with its price.
  • Independent scores on 70 models come from my hand checked rankings, each with the source and the date it was checked. A blank means no independent board has measured it yet, never a guess.
  • One entry per model. Dated snapshots and the same model listed by dozens of hosts fold into one page, so the list shows models, not endpoints.
  • Summaries on each model page are written from the facts above, so every sentence can be checked against the numbers next to it.

Choosing a model, answered

How many AI models can I compare here?

900 models from 72 labs, refreshed every day: language models plus image, video, speech, transcription, embedding and rerank models. 70 of them also carry independent scores I check by hand, and 405 have open weights you can host yourself. The newest is Nemotron 3.5 Super VL 120B A12B from NVIDIA, released Oct 15, 2026.

What is the cheapest capable AI model API?

Among models with tool calling and independent scores, Command R7B is the cheapest at $0.0375 per million input tokens and $0.15 per million output tokens. Sort the directory by price and tick "Independently scored" to see the rest of the shortlist.

Which AI model has the longest context window?

Qwen Long 10M accepts 10M tokens of input. A long window is useful for whole codebases or contracts, but price and accuracy both drop as prompts grow, so retrieval is often cheaper than filling the window.

What is the difference between input and output token prices?

Input tokens are what you send (instructions, documents, chat history). Output tokens are what the model writes back, and they usually cost three to eight times more. Agents and RAG apps send far more than they receive, so the input price and the cache discount often decide the bill.

Where does the data come from?

Prices, context windows, modalities, capabilities and the list of providers come from models.dev, an open, MIT licensed database of AI models and the hosts that serve them, and refresh daily (last on Oct 15, 2026). Independent scores come from Artificial Analysis, Arena, Terminal-Bench, ARC Prize and Scale Labs, each with its source on the model page. Descriptions are written from those facts, not copied.

How do I choose a model for an AI agent?

Filter for tool calling and structured output, set the context you need, then sort by price. Shortlist two or three, and test them on your real tasks. Most agents I ship use a cheap fast model for routine steps and a stronger one only where it pays.

A shortlist is not a decision. Your own tasks are.

I test your real workload across the two or three models that fit, then ship the agent on the one that wins on cost and accuracy, with the code in your repository.

Book Meeting