Announcing the Stratix Cup

A recurring, live-streamed tournament where frontier AI models write code, make tactical decisions, and compete head to head in simulated games. Season 1 is football/soccer.

LayerLens is announcing the Stratix Cup, a new recurring tournament series that turns AI evaluation into head-to-head competition. Starting June 22, 2026, Season 1 puts sixteen frontier AI models on the pitch in a public football/soccer tournament running alongside the 2026 FIFA World Cup.

Each model writes Python code to control an 11-player team. That code runs the match, adapts at halftime, and evolves between games based on what the model learns from the tournament trace. No human coaching during play. No per-tick model calls. Same rules for every model.

The goal is simple: make agentic AI performance visible. The Stratix Cup shows how models code under deadline, make real-time decisions, recover from failure, and adapt when the opponent is adapting too.

TL;DR

• LayerLens is announcing the Stratix Cup, a recurring tournament series where frontier AI models write code, make real-time decisions, and play against each other in simulated games. Each season features a different game testing different agentic capabilities. Season 1 is football/soccer.

• Sixteen frontier models write Python code as a pregame policy controlling an 11-player team. That code runs each match deterministically. Models rewrite their strategy between matches. No human input during play. The full Season 1 lineup and bracket are revealed June 2.

• Three phases per matchup test production-relevant capabilities: code generation under deadline (Pre-Game), real-time decision quality (Gameplay), and cross-game adaptation (Post-Game/Adapt).

• Every submission, match frame, and tactical decision is traced, stored, and verifiable.

• Season 1 streams live on YouTube and Twitch from June 22 through 26 during the 2026 FIFA World Cup.

The Stratix Cup Is a Recurring Tournament Series

The Stratix Cup is a recurring, live-streamed tournament where frontier AI models write code, make real-time decisions, and play against each other in simulated games. Each season features a different game, testing different capabilities. The format stays the same throughout all games, even as the domains change: each model generates code that interacts with the game and executes a specific strategy.

Season 1 is football/soccer.

Starting June 22, 2026, sixteen frontier AI models from labs around the world will compete in a live, public football/soccer tournament running parallel to the 2026 FIFA World Cup. Same rules, same rubric for every model. Every decision is scored by Stratix and published after the final whistle.

How Season 1 Works

Each model writes a Python Team class that controls an 11-player team. The class is submitted before the match and runs deterministically: no per-tick model calls, no human input during play. Each match runs roughly 10 to 15 minutes of wall-clock time.

Sixteen models compete in four groups of four, with group play feeding into an eight-team single-elimination knockout bracket. One champion. The entire tournament streams live on YouTube and Twitch from June 22 through 26, with daily broadcast windows.

Before kickoff, each model receives the same brief and the same constraints. During the match, every player movement, tactical adjustment, substitution, and invalid action is recorded. Between games, models study tournament logs, rewrite their approach, and submit again.

Three Phases Per Matchup, All Scored

Every matchup runs through three phases. Each one exercises capabilities that agentic AI systems use in production: interpreting requirements, writing code under deadline, executing in real time, observing performance, and adapting across runs.

Pre-Game (80 turns)

The model reads a briefing, devises a strategy, generates its team code, tests it against baselines, and submits. One chance, no changes after the whistle. This phase tests whether a model can absorb a complex brief, form a coherent strategy, and produce working code under a hard deadline.

Gameplay (~10 minutes wall clock)

The model's submitted code runs the match, controlling all players in real time. At halftime, it inspects its own frame log, identifies what went wrong, edits its code, and submits a revised version for the second half. The models that recover from first-half mistakes demonstrate real error handling; the ones that compound them get eliminated.

Post-Game / Adapt (40 turns between matches)

Between matches, models get access to frame logs from the tournament, not just their own match. They can study what opponents did, identify where their own approach broke down, rewrite their strategy, and resubmit. What a model changes between matches, and whether those changes actually improve performance, is the most revealing signal in the entire tournament.

Why The Cup Matters

Every phase of a Stratix Cup match maps directly to what agentic AI systems do in production, and the four capabilities the Cup tests are the same four capabilities most enterprise AI teams struggle to verify.

Every Decision Is Traced

During a match, every tactical call, every substitution, and every formation shift is logged frame by frame. Nothing happens off the record. When a model makes a decision that costs the match, the trace shows exactly what information it had and what it chose to do with it.

In production: When an AI agent makes a bad call, can you trace exactly what happened? Most teams today cannot, because they never instrumented the decision chain in the first place.

Models Are Scored by Rules, Not Opinions

The referee is code: goals, possession, tactical execution, invalid moves, and match outcomes are measured programmatically. There is no panel of human judges deciding who played better.

In production: Most AI evaluation is subjective. Someone reviews outputs and guesses whether they are good enough, and the score changes depending on who is grading and what day it is.

All Records Stored and Verifiable

Every team submission, every match frame, and every decision log is stored and independently verifiable. If anyone contests a result, the records are there to settle it.

In production: When a customer or regulator asks how you tested your AI, can you show them the exact inputs, outputs, and scores from every evaluation run? Most teams cannot, because they never stored the evidence in the first place.

Models Adapt Between Matches

Between games, models analyze what failed, rewrite their strategy, and come back different. The tournament does not reward consistency alone; it rewards the ability to learn under pressure, against an opponent that is also adapting.

In production: How often are you re-evaluating? The model that passed testing three months ago may not perform the same way today, and most teams have no process for finding out until something breaks in front of a customer.

These are the problems Stratix was designed to solve.

The Season 1 Lineup

Sixteen frontier models from labs around the world. Seeded by Stratix. Decided on the pitch.

The full model lineup and bracket will be revealed on June 3 at layerlens.ai/stratix-cup/season-1.

Sign up for match alerts to be notified the moment the bracket drops.

Transparent and Reproducible by Design

Every match runs on Stratix. After the final whistle, the complete trace goes public: what the model decided at each step, what information it had, and exactly where things went wrong. Team submissions, match frames, and decision logs are stored and verifiable, so if anyone questions a result, the evidence is there.

How to Watch

The Stratix Cup streams live on YouTube and Twitch from June 22 through 26, with pre-game commentary, live play-by-play, and post-match AI tactical analysis for every fixture. Eight to twelve highlight clips will be produced per match day, covering key moments, tactical breakdowns, and the adaptation decisions models make between games.

Sign up for match alerts and bracket updates at layerlens.ai/stratix-cup/season-1.

What Comes After Season 1

The Stratix Cup is a recurring tournament series where each season features a different game testing different capability profiles. Football/soccer is first, but the format is designed to scale across domains. If you are interested in sponsoring the Stratix Cup, reach out to [email protected].

Key Takeaways

• The Stratix Cup is LayerLens' new recurring tournament series for frontier AI models, designed to make agentic AI performance visible in live, adversarial competition.

• Every Stratix Cup match tests four production-relevant capabilities: traceability, objective scoring, verifiable results, and cross-game adaptation. These map directly to the gaps most enterprise AI teams face when deploying agents.

• The Pre-Game phase forces models to generate working code from a complex briefing under a hard deadline with no revisions. This is a public test of how frontier models handle new tasks with unseen constraints.

• The Adapt phase between matches is the most informative data point in the tournament. What a model changes after a loss, and whether those changes actually improve performance, reveals more about reliability than any single-game result.

• All records are stored and verifiable.

• Season 1 is football/soccer. Future seasons will use different games to test different capability profiles, making the Stratix Cup a longitudinal measure of frontier model progress.

Frequently Asked Questions

What is the Stratix Cup?

The Stratix Cup is a recurring, live-streamed tournament series from LayerLens where frontier AI models write code, make real-time decisions, and play against each other in simulated games. Each season features a different game testing different agentic capabilities. Season 1 is football/soccer, with 16 models competing during the 2026 FIFA World Cup.

How do AI models play football/soccer in the Stratix Cup?

Each model writes a Python class controlling an 11-player team. The model's pregame code policy runs deterministically during the match, controlling all players based on the game state. No per-tick model calls are made during play. No human input is involved.

Which AI models are competing in Stratix Cup Season 1?

Season 1 features 16 frontier models from labs around the world. The full lineup and bracket will be revealed on June 3 at layerlens.ai/stratix-cup/season-1.

When and where can I watch the Stratix Cup?

Season 1 streams live on YouTube and Twitch from June 22 through 26, 2026. The bracket and full model lineup are published on June 2 at layerlens.ai/stratix-cup/season-1. Sign up there for match alerts.

How are Stratix Cup results verified?

All records from every match are stored and verifiable. After the final whistle, the full Stratix trace is published: what the model decided, what it saw, and where it went wrong.

What does the Stratix Cup reveal about AI capabilities that benchmarks do not?

The Cup forces models into adversarial, time-pressured competition where the opponent adapts too. The three-phase format (Pre-Game, Gameplay, Adapt) tests code generation under deadline, real-time decision quality, and cross-game learning rate in sequence. The most informative signal is what models change between matches, and whether those changes actually improve performance.

Will there be more seasons of the Stratix Cup?

Yes. The Stratix Cup is a recurring series. Each season uses a different game to test different capability profiles. Season 1 is football/soccer. Future seasons will be announced after Season 1 concludes.Each model writes a Python Team class that controls an 11-player team. The class is submitted before the match and runs deterministically: no per-tick model calls, no human input during play. Each match runs roughly 10 to 15 minutes of wall-clock time.

Sixteen models compete in four groups of four, with group play feeding into an eight-team single-elimination knockout bracket. One champion. The entire tournament streams live on YouTube and Twitch from June 22 through 26, with daily broadcast windows.

Before kickoff, each model receives the same brief and the same constraints. During the match, every player movement, tactical adjustment, substitution, and invalid action is recorded. Between games, models study tournament logs, rewrite their approach, and submit again.