The European executive reasoning benchmark

EuroExec measures whether frontier models can make the decisions a European C-suite is accountable for, across finance, business, product, and marketing. Every task is authored and blind-scored by domain experts against a written rubric.

Every result below comes from 598 double-judged tasks: six frontier models, each answer graded blind by two independent experts.

Request dataset access Read the methodology

Leaderboard

44 vetted domain experts authored the tasks across finance, business, product, and marketing, each with a written grading checklist; 47 independent experts judged them. Six frontier models answer every task blind, and 598 tasks are fully double-judged.

The headline metric is Reward: the share of tasks a model solves, where a solved task scores a mean rubric of at least 3.0/5 and covers at least 60% of the checklist.

Explore the full leaderboard
01 Fable 5
58%
02 GPT-5.5
52%
03 Claude Opus 4.8
36%
04 Gemini 3.1 Pro
27%
05 GLM-5.2
23%
06 Mistral Large 3
19%

Difficulty

Across every blind-graded response:

28%
of tasks are unsolved by every model
51%
mean checklist coverage
94%
of tasks never fully covered by any model
28%
of answers score ≤ 2/5 on Local / Regulatory Fidelity
26%
solve rate in Marketing, the hardest domain

Explore the dataset

Five pages open the benchmark up in full, from how each task is constructed to a single graded task you can read from prompt to final scores.

Methodology

How it's built

The seven-stage pipeline from an expert's scenario to a double-judged gold record.

Open → Dataset

Composition & reliability

Domain mix, the rubrics, checklist depth, and how much two independent experts agree.

Open → Leaderboard

Full model performance

Per-rubric scores, blind rank distribution, and the head-to-head matchup grid.

Open → Failure Modes

Where models fail

Expert-written failure modes, counted by category for every model in the lineup.

Open → Sample Task

One task, end to end

An executive scenario, six blind answers, and two experts' scores, blind to revealed.

Open →

Grounded in real European markets

Every task is set in a specific European market, so a correct answer has to reason about that country's regulation and competitive reality, not a generic "Europe". The map shades each market by the share of tasks that mention it.

Germany 36.8%
Spain 27.9%
France 26.6%
Italy 19.6%
Portugal 12.9%

and 20 more markets, shaded by how often they appear. Only direct mentions count — tasks saying “Nordics” or “DACH” are not attributed to individual countries.

2% 26.6% 36.8% 0.3% 19.6% 1.5% 1.2% 12.9% 27.9% 3.2% 9.9%

Access the dataset, or contribute to it

The headline results are public. The complete task-level dataset, and a role in its expansion, are open to two audiences.

FRONTIER LABS

Full task-level results

The complete per-task breakdown across all 598 tasks and 6 models, plus private evaluation of your unreleased models on the same gold set, on request.

DOMAIN EXPERTS

Author the next batch of tasks

Senior operators in finance, business, product & tech, and marketing: author and grade the scenarios the benchmark's expansion is built on.

Request dataset access

Please fill in all fields with a valid work email.

We use your details to share the results and follow up about them. No lists, no spam. Access follows from marcus@sovrano.ai within one business day.

● REQUEST RECEIVED

You're on the list

A reply follows from marcus@sovrano.ai within one business day. What happens next:

  • Labs: the full task-level results across all 598 tasks and 6 models, plus a private eval of your unreleased models on request.
  • Domain experts: a short note on the authoring process and the domains we're expanding.

The benchmark's next release expands the domains and the language coverage. Requests hear about it first.