EQS AI Benchmark Report Vol. 2

Vol. 2 updates the industry’s first comprehensive AI benchmark for Compliance & Ethics work, now covering the latest frontier models from OpenAI, Google, Anthropic and Mistral — tested against the same 120 real-world tasks used in Vol. 1.

A few things stood out in this year’s results. The new generation made its biggest gains in open-ended Compliance work — policy drafts, investigation plans, board-level reports — improving by up to 18 percentage points compared to last year’s models. Five models from three vendors now sit within less than one percentage point of each other at the top of the leaderboard. And for the first time, a single frontier model carries a full conflict of interest workflow end-to-end, above 90%.

The report also covers where human oversight remains essential, what the convergence at the top of the leaderboard means for tool selection, and what practical steps Compliance teams can take now.