Financial evaluations - Gina

We have run financial evaluations across multiple models on tasks spanning spot markets, perpetual futures, and prediction markets. Each attempt is graded against task-specific checks for tool use and task completion. The results are published on Gina Evals.

What we evaluate

The tasks test whether a model selects the required tools, supplies the required arguments, follows the task’s restrictions, and completes without errors. Some tasks need a single tool call; others require several calls in a particular order. Each attempt starts a fresh conversation with access to read-only financial tools.

Explore the results

Model leaderboard: compare pass rates across task categories, with timing and estimated cost where available.
Tasks and grading criteria: inspect the prompts and checks used to grade each attempt.
Evaluation methodology: see how runs are conducted, scores are calculated, and missing data is handled.

The published scores measure tool use and task completion; final-answer accuracy and trading returns are not scored. Models used different clients, and the samples are small, so read the methodology and run details alongside the results. To try Gina with your own assistant, visit Get Started.