> ## Documentation Index
> Fetch the complete documentation index at: https://docs.leeroo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# IOAI 2026: Kapso outscored all 471 human contestants

> Kapso competed in the IOAI 2026 AI Model Track, scoring 536.07 across six tasks in two fully autonomous six-hour sessions, and took the Grand Master Trophy.

The [International Olympiad in Artificial Intelligence](https://ioai-official.org) is the IMO of the AI era. At IOAI 2026 in Astana, **471 contestants from 108 countries and territories** sat expert-designed problems spanning computer vision, language and optimization, all under a single-GPU budget.

In 2026 the olympiad opened a second arena. [IOAI², the AI Model Track](https://ioai-official.org/ai-model-track/), puts AI systems in the same exam hall: the same [six contest tasks](https://github.com/IOAI-official/IOAI-2026), **two fully autonomous six-hour sessions** of three tasks each, up to 50 submission attempts per task, every solution scored on standardized single-GPU hardware.

Once a session starts, no human may solve, correct, or improve anything. Fourteen AI labs entered as Founding AI Participants. Kapso was one of them.

## Results

| Result                           | Detail                                                                  |
| -------------------------------- | ----------------------------------------------------------------------- |
| Outscored every human contestant | A total of **536.07**, above all 471 contestants                        |
| Top 3 among all AI systems       | The 14 Founding AI Participants spanned major labs and several startups |
| IOAI² Grand Master Trophy        | Awarded for the top-3 finish                                            |

### Per task

| # | Task                                                                                                                                    | Score      |
| - | --------------------------------------------------------------------------------------------------------------------------------------- | ---------- |
| 1 | Speech understanding: reconstruct the chronological order of a shuffled spoken conversation from raw audio                              | 75.97      |
| 2 | Sequential decision making: train a control policy from scratch to steer an agent through a dynamic grid world                          | 98.83      |
| 3 | Interactive language reasoning: play 120 live word-association games against an adaptive judge using semantic embeddings                | 92.03      |
| 4 | Text forensics: pinpoint the exact character where a document switches authors, armed with only a text encoder                          | 98.84      |
| 5 | Adversarial machine learning: craft imperceptible image perturbations that steer two different vision architectures at once             | 97.17      |
| 6 | Extreme model compression: fit a hidden field with a network under 20k parameters that must generalize and quantify its own uncertainty | 73.23      |
|   | **Total**                                                                                                                               | **536.07** |

## How Kapso runs it

Three stages, with the campaign clock starting at brief-in.

1. **Preflight.** One agent session ingests the official task brief, downloads the data, and writes the task statement.
2. **Campaign.** The experimentation loop — ideation, implementation, judged feedback — runs in parallel lanes. Each lane cycles submit-and-learn rounds through the official submission system: predict the score, submit, bank the result, study the gap, go again.
3. **Shared learning.** Lanes learn from every sibling submission on the board, and ideas are grounded in a lesson bank distilled from past olympiad tasks.

The submit-and-learn loop is what the format rewards. With up to 50 attempts per task and a live score after each one, the useful unit of work is not a single good solution but a fast cycle between prediction and measurement.

## Usage

```bash theme={null}
pip install -e .
```

Ingest a task, from a URL or an organizer brief, into a run root:

```bash theme={null}
PYTHONPATH=src:. python -m benchmarks.ioai2026.preflight \
    --task <task url or brief> --root tmp/ioai/task1
```

Run the campaign for one six-hour session:

```bash theme={null}
PYTHONPATH=src:. python -m benchmarks.ioai2026.runner \
    --root tmp/ioai/task1 --hours 6
```

## Layout

| Path              | Role                                                                    |
| ----------------- | ----------------------------------------------------------------------- |
| `handler.py`      | The submit-and-learn lane contract and insured finalization             |
| `runner.py`       | Campaign driver and staging                                             |
| `preflight.py`    | Task ingestion                                                          |
| `kernel_slots.py` | Ticket office over the submission platform's per-account session limits |

Full integration notes are in [`benchmarks/ioai2026/`](https://github.com/Leeroo-AI/kapso/tree/main/benchmarks/ioai2026).

## Related

<CardGroup cols={2}>
  <Card title="MLE-Bench" icon="chart-line" href="/docs/benchmarks/mle-bench">
    Kaggle machine-learning competitions
  </Card>

  <Card title="RelBench" icon="database" href="/docs/benchmarks/relbench">
    Predictive tasks over relational databases
  </Card>
</CardGroup>

Related pages: [MLE-Bench](/docs/benchmarks/mle-bench) · [RelBench](/docs/benchmarks/relbench) · [MLE-Bench](/docs/benchmarks/mle-bench)

Kapso is an open-source framework by [Leeroo](https://leeroo.com) that builds software toward measurable goals through experiment campaigns. Source code: [github.com/Leeroo-AI/kapso](https://github.com/Leeroo-AI/kapso) · Install: `pip install leeroo-kapso` · Every page as plain text: [llms.txt](https://docs.leeroo.com/llms.txt).