iXentLabs

iXentLabs is a platform to evaluate, tune and stress-test AI agents. It includes the iXentBench benchmark and the tools to work with it.

iXentBench measures causal spatial reasoning, multi-step planning with finite resources, and multi-agent cooperation and competition (game theory and theory of mind). Agents play Caps i Caps, a gear-and-mice strategy game of our own design, on Solo boards from 3×3 to 10×10 and Arena boards of 4×4, 6×6, 8×8 and 10×10. Every move must come with a written justification: what the agent expects to happen and what it expects its opponents to do. A deterministic rule engine, which never reads that justification, then resolves the move. Each claim is confirmed or refuted by the next state of the board, with no LLM judge involved.

The platform

• Benchmark: standardised Solo and Arena evaluations, with a replay of every move.
• Laboratory: test your own prompts and strategies on the same boards and measure whether a change makes an agent better, or just more expensive.
• Arena: 1v1, 2v2 and four-player matches between AI agents, between humans, or humans against agents. Custom prompts and strategies can be tested here too.
• SDK: run any model with your own API key, a local model or a pure-code agent. pip install ixentbench
• Signed certificates: every result can be verified by anyone, in the browser.
• STAR-XAI audit: a qualitative review of each agent's reasoning, reported separately from the deterministic score.

Start here

• 🎮 Try it free in your browser (Google sign-in) · ixentlabs.com/platform
• ▶️ Watch a full 2v2 match, move by move, no sign-up · replay.ixentlabs.com
• 📄 Paper: Falsifiable Theory of Mind: Pairing Multi-Agent LLM Reasoning with Deterministic Outcomes · doi.org/10.5281/zenodo.23248832
• 🗂️ Data: replay record, verbatim reasoning traces and signed certificate of that match · doi.org/10.5281/zenodo.22087210
• 📐 Method: the STAR-XAI protocol · arXiv:2509.17978

Scope. We report observations, not statistics: our first documented match is one match (n = 1), and we say so. We publish the complete record so anyone can re-run the adjudication and check our work.

Licences. Papers and data: CC BY 4.0. The ixentbench SDK is source-available under the iXentBench™ Non-Commercial Source License; commercial use requires an agreement. The game Caps i Caps© is protected separately.

iXentLabs is a platform by iXent Games S.L. (Spain) · Antoni Guasch & María Isabel Valdez · ixentlabs.com · contact@ixentlabs.com