What it is
LinkedIn Labs Crosscheck lets members run the same prompt against two AI models, vote on the better answer, and contribute to a segmented model leaderboard.
Gabriel’s notes
Quick take: Crosscheck is LinkedIn Labs’ “A/B test for LLMs,” aimed at answering a practical question: which model works best for the work I actually do? You submit a prompt, get two model answers, pick the better one, and your vote helps build an AI model leaderboard that can be filtered by professional context (like role and industry). ([linkedin.com](https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world))
At launch, LinkedIn described Crosscheck as available to Premium subscribers in the United States, with expansion to all U.S. members during the week of May 21, 2026, and a broader global rollout planned. ([linkedin.com](https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world)) The Labs URL currently redirects to an “ai-trainer” path on LinkedIn. ([linkedin.com](https://www.linkedin.com/labs/crosscheck))
I saved this under AI because it’s one of the rare “benchmark-ish” tools that’s designed for messy, human, real-world work (instead of benchmark cosplay).
What you described (cleaned up): The Crosscheck flow lets you send one prompt to multiple AI models (presented as a two-response “battle”), compare outputs without brand bias, and rate which answer you prefer. Those votes roll up into a public-facing leaderboard at /labs/crosscheck/leaderboard. I also think both the battle flow and the leaderboard are genuinely useful—and, yes, kind of fascinating to watch. ([linkedin.com](https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world))
Good fit if you want to:
- Do fast “vibe checks” across frontier models for the same prompt (writing, summarization, reasoning, etc.). ([linkedin.com](https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world))
- Reduce brand bias by choosing between answers before you know which vendor produced what (“blind taste test”). ([engadget.com](https://www.engadget.com/ai/linkedins-crosscheck-feature-lets-premium-subscribers-test-ai-models-for-free-183949210.html?utm_source=openai))
- Get a directional signal on which models your professional peers prefer for specific kinds of work (role/industry segmentation). ([linkedin.com](https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world))
- Pressure-test prompts you plan to reuse in a workflow, especially if you’re model-agnostic or provider-agnostic.
- Teach teams a healthy habit: “compare outputs, then decide,” instead of “pick a model and emotionally commit.”
Pricing snapshot (auto-enriched):
LinkedIn positioned Crosscheck as included with LinkedIn Premium in the U.S. at launch. ([engadget.com](https://www.engadget.com/ai/linkedins-crosscheck-feature-lets-premium-subscribers-test-ai-models-for-free-183949210.html?utm_source=openai)) LinkedIn engineering also stated it would expand to all U.S. members during the week of May 21, 2026. ([linkedin.com](https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world)) Current availability outside the U.S. (and whether it still requires Premium) is Unknown / not confirmed.
Work-use / compliance snapshot (auto-enriched):
If you’re using Crosscheck for work, assume your prompts may be logged/processed as part of operating and improving LinkedIn’s services unless a specific policy says otherwise (so: don’t paste secrets). LinkedIn provides a member setting called Data for Generative AI Improvement that controls whether your data is used to improve generative AI models used for content creation, where applicable settings/law allow. ([linkedin.com](https://www.linkedin.com/help/linkedin/answer/a6278444?utm_source=openai)) LinkedIn’s privacy policy summary also states it may use personal data to develop and train AI models. ([linkedin.com](https://www.linkedin.com/legal/privacy-policy-summary?utm_source=openai)) Whether Crosscheck battles/prompts are used to train any specific model is Unknown / not confirmed.
Alternatives (auto-enriched):
- consens.io — More of a “ask multiple models and compare” app for individual decision-making; less about professional-segment leaderboards. ([consens.io](https://www.consens.io/?utm_source=openai))
- PromptBench — Oriented toward prompt testing/analytics across models; more systematic for teams, but not the same peer-driven leaderboard concept. ([promptbench.co](https://www.promptbench.co/?utm_source=openai))
Before you adopt it:
- Write (and reuse) a small set of “golden prompts” that reflect your actual workflows—then test them regularly as models update.
- Decide your rubric before you vote: correctness, tone, format adherence, citation behavior, safety, etc.
- If you handle sensitive data, treat Crosscheck like a public SaaS surface: sanitize inputs and keep client identifiers out.
Sources:
https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-worldhttps://www.engadget.com/ai/linkedins-crosscheck-feature-lets-premium-subscribers-test-ai-models-for-free-183949210.htmlhttps://www.linkedin.com/help/linkedin/answer/a5538339https://www.linkedin.com/help/linkedin/answer/a6278444https://www.linkedin.com/legal/privacy-policy-summary