New site pits AI models against each other on text, image, video and audio

Crun AI has announced a new tool, AI Leaderboard, where users can compare AI models side by side. The announcement is dated Oct. 15-16, 2026, as the sources give different dates.

The leaderboard covers four areas: large language models (the models behind chatbots like Claude and Gemini), video generation, image generation and audio processing.

What can you compare?

According to the press release, users can see performance benchmarks, meaning test results, as well as pricing, features and context windows. A context window is the amount of text a model can handle at once. Users can also see which types of input and output the models support, what share of tasks they complete, and how reliable they are.

Among the models mentioned are GPT-6 Astra, Claude, Gemini, Kimi and DeepSeek for language. For video, Seedance 2.5, Kling, Veo and Grok Imagine are named. For images, GPT Image 2.5, Nano Banana and Seedream are named, and for audio, Suno and Qwen3-TTS are named.

The tool targets six use cases: coding and software development, AI agents and automation, research and knowledge work, marketing and advertising, film and short drama, and e-commerce content.

According to Crun AI, the leaderboard is updated weekly with new models, changed features and price changes. That is a plan from the company.

A company with its own interests

The press release states that the best AI model depends on the task, the workflow and the budget.

Crun describes itself as a unified AI platform that provides access to many models through a single API, meaning one technical entry point. The platform has more than 100 models across video, image, audio, text and 3D. This means Crun itself sells access to the models being compared, so the leaderboard is not independent.

All of the information comes from Crun's own press release and website. No independent party has tested or confirmed the tool. The sources also do not make clear how rankings and benchmarks are calculated, or what data they are based on. For readers, this means the comparisons can be a useful guide, but they should be taken with caution.