Benchmarks got harder.They never got wider.

EuroExec measures whether frontier models can make the decisions European executives make.

Evaluate your model on the full held-out set Read the methodology
598 Double-judged tasks
44 Expert authors
47 Expert judges
6 Frontier models evaluated
28% Of tasks unsolved by every model
Round 1 results

Fable 5 leads at 58%. The field ranges from 58% to 19%.

The headline metric is the solve rate (Reward): the share of tasks a model solves. A task counts as solved when the two experts' mean rubric score is at least 3.0/5 and the answer covers at least 60% of the checklist.

Explore the full leaderboard
EuroExec Leaderboard
  1. 01 Anthropic Fable 5
    58%
  2. 02 OpenAI GPT-5.5
    52%
  3. 03 Anthropic Claude Opus 4.8
    36%
  4. 04 Google Gemini 3.1 Pro
    27%
  5. 05 Z.ai GLM-5.2
    23%
  6. 06 Mistral Mistral Large 3
    19%

The gap to expert performance is still wide.

28% Of tasks unsolved by every one of the six models
92.4% Expert score on a 66-question subsample, against 58% for the best model
94% Of tasks never fully covered by any model
28% Of all model answers score 2/5 or lower on Local / Regulatory Fidelity
26% Solve rate in Marketing, the hardest domain
Five ways in

Go deeper.

Geographic coverage

Grounded in real European markets.

Every task is set in a named market. Only direct mentions count: a task that says "Nordics" or "DACH" is left out of the country counts. Percentages add up to more than 100% because many tasks name several markets.

2% 26.6% 36.8% 0.3% 19.6% 1.5% 1.2% 12.9% 27.9% 3.2% 9.9%
Germany
36.8%
Spain
27.9%
France
26.6%
Italy
19.6%
Portugal
12.9%

Plus 20 more markets, shaded on the map by how often they appear.

Access

Put your model on EuroExec.

Two ways to run your model on EuroExec. We'll run it and send you the item-level failures. Domain experts who want to write or grade the next batch can use the same form.

Full held-out evaluation

We run your model on the full held-out set and send back task-level scores, checklist coverage and both experts' written rationale for every answer.

Private checkpoint evaluation

Run EuroExec on unreleased checkpoints before launch. Results stay private to your team. We never train on, or sell data derived from, the held-out set we score you against.

Evaluate your model

We use your details only to share results and follow up on them. We reply within one business day. See our Privacy Policy.

Help us build a better platform for experts

We use analytics to see which tasks and features work for you, and marketing cookies to measure the campaigns that bring new experts in. Nothing optional is set until you choose. Change your mind anytime in Cookie settings.

Cookie policy

Cookie settings

Pick what you are comfortable with. You can change this anytime from Cookie settings.

Strictly necessary

Login, security and storing this choice. Required for the app to work.

Always on

Product analytics Helps us most

Helps us see which parts of the platform work well and where things get stuck, so we can make it smoother for you.

Marketing

Lets us see which campaigns bring new experts to Sovrano AI, so we can keep growing the community and the projects on offer.

See the full list of cookies