The paper introduces Gauntlet, a framework that lets large language models autonomously build game-playing agents from a bare contract—just a game description, raw observation/action interface, and an empty policy file. In a single session, the model experiments with the game, compiles a standalone controller, and the resulting program is evaluated on held‑out instances without further model calls. The authors demonstrate that these compiled agents can win full‑scale games such as StarCraft II and Civilization, marking the first time a language‑agent system has achieved standalone victory in such complex titles.
arXiv:2605. 05138v2 Announce Type: replace Abstract: We evaluate an initial coding-agent system for ARC-AGI-3 in which the agent maintains an executable Python world model, verifies it against previous observations, refactors it toward simpler abstractions as a practical proxy for an MDL-like simplicity bias, and plans through the model before acting.
CivBench is an open‑source benchmark that evaluates language‑model agents in the long‑horizon, tool‑mediated game Civilization VI using the Model Context Protocol (MCP). Each episode lasts over 300 turns, generating thousands of tool calls across a 76‑tool action space, and includes a narration layer that translates visual game state into structured text. The study characterises agent behaviour across four model families, introducing Proactive Monitoring Rate (PMR) and RAG@10 as interface‑level metrics, and finds that agents often under‑monitor strategic state and fail to execute near‑term commitments despite tool access and explicit guidance.
arXiv:2607. 02255v1 Announce Type: new Abstract: Memory for a long-horizon LLM agent is a contract about what each future decision is allowed to see.
Kepler is an open‑source harness that represents hypotheses as executable world models and validates them through retrospective transition checks and conditional prediction checks. In the ARC‑AGI‑3 benchmark, a frozen Claude Opus 5 configuration achieved a perfect 100.00 RHAE on all 25 public games without per‑game model selection or score‑conditioned reruns, and matched or outperformed median‑human action counts on 181 of 183 levels. The study also identified three evaluation failures and highlighted that public‑set score alone has limited discriminative value, advocating for first‑attempt, cost‑conditioned, and verification‑aware reporting. whyItMatters":"The results demonstrate that a purely score‑based evaluation can be misleading, underscoring the need for more rigorous, cost‑aware, and verification‑aware metrics in AI benchmark assessments."
SpireBench evaluates long‑horizon strategic play by having large language models play the full commercial deck‑builder Slay the Spire, a game where early choices determine long‑term success. Across 20 LLMs at the highest difficulty, frontier models beat heuristic baselines and match average human performance but lag far behind experts, especially as runs lengthen. The benchmark reveals that strong models excel in early fights but misjudge risk over a run, and that their performance depends partly on prior game knowledge rather than on memories written during play.
arXiv:2606. 16014v1 Announce Type: cross Abstract: Many games rely on storytelling combined with systems that track levelling, NPC behaviour, and consequence simulation; bridging tightly-authored narrative with deeply-simulated worlds -- most acute in sandbox and open-world settings -- has been prohibitively expensive.
The paper introduces GUI agents for continual game generation, presenting PlaytestArena—a benchmark of 200 browser-based game-generation tasks with rubrics for in‑play behavior—and Play2Code, an iterative framework where a game agent and a rubric‑blind GUI playtester refine games through shared memory. Play2Code achieves a 66.8% rubric pass rate, surpassing baseline methods by 37.1 and 14.6 points, and shows consistent score improvement across refinement rounds. The study demonstrates that GUI playtesting provides actionable, traceable feedback that can guide interactive code generation.
The paper introduces CivHarness, a system that records game states in Freeciv and replays them to generate long‑horizon reward signals, addressing the problem of sparse deterministic feedback in strategy games. Using CivHarness, the authors train CivTelescope, an LLM judge that predicts which of two positions will lead to a better outcome, outperforming standard language models on scoreboard‑trap pairs. They then employ CivTelescope as a dense reward in a Freeciv agent, CivMarshmallow, achieving significantly higher training gains and outperforming other reward schemes across the full game.
arXiv:2609.21562v1 Announce Type: cross Abstract: Coding agents can modify and test code across large software projects. Game development is a domain where agents must implement gameplay rules. A gam...
arXiv:2608.21833v1 Announce Type: new Abstract: Recent large language models (LLMs) can operate as coding agents that build complete games from natural language requests. Game development is especial...
arXiv:2610.33845v1 Announce Type: new Abstract: Can LLM agents carry a hard real-world problem from raw data to a system that holds its own against people? We study this in Hearthstone Battlegrounds,...
The paper introduces Skillware, an open architectural framework that separates an agent’s intent from its execution by providing modular, versioned software capability bundles. These bundles include typed JSON contracts, deterministic Python code, task directives, datasets, and unit tests, allowing agents to perform tasks reliably without relying on fragile prompt engineering. Experiments across 23 models and 368 benchmarks show that small local models equipped with Skillware outperform larger frontier models, reducing token usage by up to 84.8%, cutting inference costs by up to 97.3%, and eliminating parameter errors.