LinkedIn launches Crosscheck: blind comparison testing for AI models from OpenAI, Anthropic, Google

LinkedIn Premium subscribers in the United States can now test and compare AI models from OpenAI, Anthropic, Google, Mistral, MoonshotAI, and Amazon through a new feature called Crosscheck, with no token limits or additional subscription costs.

How it works

Crosscheck presents users with a single prompt and two responses from different AI models. Users select their preferred response before seeing which models generated each answer. According to LinkedIn Chief Product Officer Hari Srinivasan, the feature is designed as a "blind taste test" for AI models.

The feature includes a leaderboard tracking model preferences across different professional industries.

Current limitations

Crosscheck supports text-based prompts only. Users cannot generate images, upload files, or access advanced features available on native AI platforms. However, there are no limits on the number of text-based conversations.

Srinivasan describes Crosscheck as an "early product" from LinkedIn Labs, noting "there's work to do to make it faster and add more models and question types."

Data sharing policy

LinkedIn shares anonymized usage data with model providers. According to the company's documentation, "Anonymized data is shared with model builders to help them understand how their models are performing amongst different occupations. No personally identifiable information is shared with model builders."

Availability

Crosscheck is currently limited to LinkedIn Premium subscribers in the United States. LinkedIn plans to expand the feature to additional countries and free users "soon," though no specific timeline was provided.

What this means

LinkedIn's move positions the professional network as a neutral testing ground for AI models, potentially influencing enterprise adoption decisions based on professional use cases. The blind testing format and occupational data collection could provide model providers with valuable insights into industry-specific performance, while the token-free access removes a significant barrier for professionals exploring AI tools. However, the data-sharing arrangement means users are effectively providing training feedback to multiple AI companies simultaneously.