Tycho: Active Abstraction with Programmatic World Models for ARC-AGI-3

Source

Source name: arXiv
Source kind: paper
Authors / organization: Jens Lehmann, Andrei Aioanei, Sahar Vahdati
Venue: arXiv: cs.AI
Published: 2026-07-30
Collected: 2026-07-31
URL: http://arxiv.org/abs/2607.28287v1
PDF: https://arxiv.org/pdf/2607.28287v1
DOI: UNKNOWN

Plain-Language Brief

The paper introduces Tycho, a system that helps an artificial agent learn the rules and goals of unfamiliar games. The system separates actionable observations from other frames and uses them to construct game-specific models during interaction. Four orchestration policies were compared using Claude Opus 4.8 under matched inference budgets. Actor-requested delegation to a model builder resulted in the highest observed mean Relative Human Action Efficiency (RHAE).

What the Document Says

ARC-AGI-3 turns abstraction into an interactive problem of skill acquisition.
The paper introduces Tycho, a coding-agent system that constructs and uses game-specific models during interaction.
Tycho separates actionable observations from intermediate animation, level-completion, and game-over frames.
From this structured history, an agent can model, test, plan with, repair, or bypass a free-form executable hypothesis.
Four orchestration policies were compared on all 25 public games using Claude Opus 4.8 under matched inference budgets.
Actor-requested delegation to a model builder obtained the highest observed mean Relative Human Action Efficiency (RHAE), 88.49.
With the selected policy, GPT-5.6 Sol and Opus 5 both reached 100.00 RHAE and completed all 183 levels.
Their game-balanced first-run human-replay midranks were 98.5 and 100.0.
Opus 5 used 61% fewer scored actions than the aggregate official human baselines.
Automatic repair after verification failures produces models that reproduce observed transitions more accurately, yet reaches only 83.07 RHAE.
Strong play also requires deciding when to construct, repair, use, or bypass a model. This joint problem is called active abstraction.

Key Concepts Mentioned

ARC-AGI-3
Tycho
abstraction
game-specific models
orchestration policies
Claude Opus 4.8
Relative Human Action Efficiency (RHAE)
GPT-5.6 Sol
Opus 5

Claims Made by the Document

Actor-requested delegation to a model builder obtained the highest observed mean RHAE.
Opus 5 used fewer scored actions than official human baselines.

Evidence Used

In one matched public-set run per policy, we compare four orchestration policies on all 25 public games using Claude Opus 4.8 under matched inference budgets. Actor-requested delegation to a model builder obtains the highest observed mean Relative Human Action Efficiency (RHAE), 88.49.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

ARC-AGI-3 turns abstraction into an interactive problem of skill acquisition. A player must infer an unfamiliar game's rules, hidden state, and goal while maintaining action efficiency because every move counts. We formalize these environments as parameterized rendered deterministic Moore machines and introduce Tycho, a coding-agent system that constructs and uses game-specific models during interaction. Tycho separates actionable observations from intermediate animation, level-completion, and game-over frames. From this structured history, an agent can model, test, plan with, repair, or bypass a free-form executable hypothesis. In one matched public-set run per policy, we compare four orchestration policies on all 25 public games using Claude Opus 4.8 under matched inference budgets. Actor-requested delegation to a model builder obtains the highest observed mean Relative Human Action Efficiency (RHAE), 88.49. With this selected policy, GPT-5.6 Sol and Opus 5 both reach 100.00 RHAE and complete all 183 levels. Their game-balanced first-run human-replay midranks are 98.5 and 100.0. Opus 5 uses 61% fewer scored actions than the aggregate official human baselines. Automatic repair after verification failures produces models that reproduce observed transitions much more accurately, yet reaches only 83.07 RHAE. Transition match indicates whether a simulator reproduces observed dynamics, not whether it has identified the objective or improves the next action. Strong play also requires deciding when to construct, repair, use, or bypass a model. We call this joint problem active abstraction: generating a testable model from costly interaction and deciding when acquiring or using it is worth its cost.