OpenAI의 GPT-6 Astra가 포켓몬 파이어레드를 18시간 만에 클리어하는 등 이전 모델들이 실패한 게임들을 잇달아 정복했습니다. 특히 마인크래프트에서는 엔더의 눈 재료까지 모았으나 크리퍼 폭발로 창고가 날아간 뒤 몇 시간 동안 감자 농사만 짓는 해프닝도 벌어졌습니다. ARC-AGI-3 벤치마크에서도 62.7%를 기록하며 GPT-5.6 대비 큰 도약을 보여주었습니다.
번역된 본문
GPT-6 Astra: 18시간 만에 포켓몬 챔피언, 크리퍼 사고 한 번으로 감자 농부가 되다
96시간이 아닌 18시간 만에 포켓몬 챔피언 등극, 팩토리오에서 로켓 발사, 폴아웃 3 엔딩 크레딧까지. GPT-6 Astra는 이전의 모든 모델을 궁지에 몰아넣었던 게임들을 정복했다. 하지만 크리퍼가 보관함을 날려버린 뒤에는 마인크래프트에서 몇 시간 동안 감자 농사만 지었다.
블레이즈 막대 6개, 엔더 진주 3개. Vals AI에 따르면, GPT-6 Astra만큼 마인크래프트에서 이만큼 진전을 이룬 AI 시스템은 없었다. Astra는 네더(지옥)에서 반자동 블레이즈 농장을 짓고, 뒤틀린 숲에서 엔더맨 여섯 마리 이상을 처치한 뒤, 게임의 최종 결전으로 가는 포탈을 찾을 수 있게 해주는 엔더의 눈 재료를 모았다. Astra가 전리품을 상자에 넣을 때 수천 명이 스트림을 지켜봤다. 그리고 크리퍼가 나타났다. 폭발로 상자와 침대가 사라졌다. 모든 것이 날아갔고, Astra가 이를 알아차렸을 때는 비까지 내리기 시작했다. Astra는 일부 대문자로 메모를 남겼다: "중요한 아이템은 항상 휴대할 것(keepInventory 사용); 보호되지 않은 상자에는 다시 절대 보관하지 말 것." 그리고 그 후 몇 시간 동안 감자를 거의 재배하는 데만 보냈다.
이전 그 어떤 모델보다 빠르게
그 전에 GPT-6 Astra는 게임에서 100시간이 넘는 시간을 보냈다. Vals AI는 결국 141시간 만에 실행을 중단했다. 초기 마인크래프트 AI 에이전트 시절과 비교하면 이 도약의 크기가 명확해진다. 2022년, OpenAI의 VPT는 실행의 겨우 2.5%에서만 다이아몬드 곡괭이에 도달했고, 포탈에는 한참 못 미쳤다. Vals AI의 한 비교에서 VPT는 거기까지 도달하는 데 10분이 걸렸다.
최근 모델과의 가장 깔끔한 비교는 다른 프로젝트인 'GPT Plays Pokemon'에서 나온다. 운영자 Clad3815에 따르면, Astra는 포켓몬 파이어레드를 18시간 12분 만에 챔피언 칭호를 얻었다. GPT-5.6 Sol은 96시간 35분이 필요했고, GPT-5.5는 218시간이 넘도록 게임을 끝내지 못했다.
'팩토리오: 스페이스 에이지'의 커뮤니티 실행도 비슷한 이야기를 들려준다. MCP 인터페이스가 있는 커스텀 루아(Lua) 모드를 통해 조작된 Astra는 약 2시간 만에 파란 과학 팩을 생산했고 대략 10시간 만에 첫 로켓을 발사했다. 같은 설정에서 GPT-5.6 Luna와 Fable 5.1은 전력 공급과 석유 탐사 단계를 넘지 못했다.
직접적인 비교 수치가 없는 클리어 사례도 있다. 포탈을 끝까지 클리어해 크레딧을 봤고, 폴아웃 2를 22시간 만에 끝냈으며, 림월드에서는 여러 차례 습격을 버텨내고 결국 행성을 벗어난 식민지를 만들었고, 폴아웃 3의 메인 스토리를 약 59시간 만에 완주했다. 커뮤니티 실행 전에도 ARC-AGI-3는 이것이 한 걸음이 아니라 도약임을 보여줬다. 이 벤치마크는 모델을 낯선 추상적 게임 환경에 던져 넣고 스스로 행동하며 규칙을 파악하게 한다. Astra는 표준 인터페이스에서 62.7%를, OpenAI 자체 하네스에서 약 99.9%를 달성했다. GPT-5.6 Sol은 7.78%, Claude Opus 5는 30% 조금 넘는 점수를 받았다.
먼저 확인하고, 그다음 반복하라
벤치마크를 운영하는 단체인 ARC Prize는 이 도약이 어떻게 일어났는지 설명했다. Astra는 낯선 게임 메커니즘을 간결한 기호적 서술로 변환하고, 물체·좌표·규칙·계획된 행동을 스스로 개발한 약식 표기로 추적한다. 관찰이 규칙이 되고, 규칙이 계획이 된다.
경험을 재사용 가능한 규칙으로 바꾸는 아이디어는 새롭지 않다. 2023년, 엔비디아와 칼텍이 참여한 연구 프로젝트 보이저(Voyager)는 GPT-4가 마인크래프트에서 과제를 제안하고, 이를 위한 자바스크립트 코드를 작성하며, 오류 메시지를 바탕으로 코드를 수정하게 했다. 작동하는 프로그램은 스킬 라이브러리에 저장됐다. 하지만 보이저는 게임을 직접 보지 못했다. 구조화된 데이터를 받고 Mineflayer 프로그래밍 인터페이스를 통해 게임을 조작했다. 반면 Vals AI의 마인크래프트 실행은 회사에 따르면 화면, 마우스, 키보드를 통한 일반적인 컴퓨터 사용 방식으로 진행되며, 별도의 전용 게임 연결이 없다. 연구자들이 예전에 모델 주변에 구축했던 것을, Astra는 이제 스스로 해내고 있다.
GPT-6 Astra: Pokemon champion in 18 hours, potato farmer after one Creeper mishap Maximilian Schreiner View the LinkedIn Profile of Maximilian Schreiner Sep 17, 2026 Nano Banana Pro prompted by THE DECODER Pokemon champion in 18 hours instead of 96, a rocket launch in Factorio, the credits rolling in Fallout 3. GPT-6 Astra beats games that stumped every model before it. But then it spent hours farming potatoes in Minecraft after a Creeper blew up its stash. Six blaze rods, three ender pearls. According to Vals AI, no AI system had ever gotten this far in Minecraft as GPT-6 Astra. In the Nether it had built a semi-automatic blaze farm, killed more than half a dozen Endermen in a warped forest, and gathered the ingredients for eyes of ender, which let you find the portal to the game's final battle. Thousands watched the stream as Astra put its loot in a chest. Then came the Creeper. The explosion wiped out the chest and the bed. Everything was gone, and by the time Astra noticed, it had started to rain too. Astra jotted down a note, partly in caps: "ALWAYS CARRY CRITICAL ITEMS with keepInventory; don't store in unguarded chest ever again." Then it spent several hours farming almost nothing but potatoes. Faster than any model before it Before that, GPT-6 Astra had spent well over 100 hours in the game. Vals AI finally called off the run after 141 hours. Compare that to the early days of the first Minecraft AI agents, and the size of the jump becomes clear. In 2022, OpenAI's VPT reached a diamond pickaxe in just 2.5 percent of runs, nowhere near a portal. In one Vals AI comparison, VPT needed 10 minutes to get there. The cleanest comparison to more recent models comes from a different project, " GPT Plays Pokemon ". According to operator Clad3815, Astra earned the champion title in Pokemon FireRed after 18 hours and 12 minutes. GPT-5.6 Sol needed 96 hours and 35 minutes. GPT-5.5 still hadn't finished the game after more than 218 hours. A community run of Factorio: Space Age tells a similar story. Controlled through a custom Lua mod with an MCP interface, Astra produced blue science packs after about two hours and launched its first rocket after roughly ten. In the same setup, GPT-5.6 Luna and Fable 5.1 never got past power supply and oil prospecting. Then there are completions with no direct comparison numbers. Portal all the way to the credits , Fallout 2 in 22 hours , a colony in RimWorld that Astra carried through several raids and eventually off the planet, plus the main story of Fallout 3 after about 59 hours. Even before the community runs, ARC-AGI-3 had already signaled that this was a jump, not just another step. The benchmark drops models into unfamiliar, abstract game environments where they have to figure out the rules through their own actions. Astra hit 62.7 percent over the standard interface, and about 99.9 percent with OpenAI's own harness. GPT-5.6 Sol scored 7.78 percent, and Claude Opus 5 a little over 30 percent. Check first, then repeat ARC Prize, the organization behind the benchmark, explained how that leap happened: Astra translates unfamiliar game mechanics into compact symbolic descriptions and tracks objects, coordinates, rules, and planned actions in a shorthand it develops itself. Observations become rules, and rules become plans. The idea of turning experience into reusable rules isn't new. Back in 2023, the research project Voyager , which involved Nvidia and Caltech, had GPT-4 propose tasks in Minecraft, write JavaScript code for them, and revise that code based on error messages. Working programs went into a skill library. But Voyager never saw the game. It got structured data and controlled the game through the Mineflayer programming interface. Vals AI's Minecraft run , by contrast, runs through general computer use, meaning screen, mouse, and keyboard, with no specialized game hookup, according to the company. What researchers once built around the model, Astra now does on its own. In under three hours it had built a Nether portal, fought zombies, trapped a skeleton, and found its way back to base. The Fallout 3 run shows a similar structure. The user imjustnewatai, who posted the full run on YouTube, played the opening in Vault 101 himself and then handed the save file to OpenAI's agent tool Codex, with Astra set to maximum reasoning. From there the agent picked its own route and actions. Galaxy News Radio, the search for the father, the G.E.C.K., the escape from Raven Rock, Project Purity. It played through normal game inputs in cycles of pausing, observing, and acting. Saving, companions, and fast travel were allowed. After about 59 hours, the credits rolled. The pattern ARC Prize described shows up most clearly in the hardest stretch of the run. In Vault 87 the character died several times, ammo ran out, and Astra had to backtrack to heal. The status messages from this phase all follow the same shape. An action has no effect, and instead of repeating it, Astra looks for the cause and records the result. A bed that was supposed to heal the character didn't respond at first. Astra let the game run for a moment, tried again, and reported: "The bed worked after letting the game run briefly. Health is now fully restored to 260/260." Shots at an enemy kept missing. Astra figured there was an obstacle in the line of fire, moved closer, and noted: "Moving closer cleared the obstruction that was spoiling the earlier shots." At a keypad, it waited until the game confirmed the button press before moving on. Looking back, Astra boiled these episodes down to a single principle. Check whether an input worked before repeating it. That sounds trivial, but it's the skill that keeps the agent from looping the same useless action over and over. It can also be the starting point for behavior that later becomes a problem. Never an unguarded chest again The same run also shows where this tips over. After the deaths in Vault 87, status messages piled up in which Astra saved its progress, checked the recording, and vetted every step against the visible environment before moving on: "I'm checking each short move against the visible geometry so I can leave without losing the saved progress." That slowed the run down a lot, but it didn't grind it to a halt. In Minecraft it did. Right after the Creeper explosion , Astra wrote itself a rule. Always carry important items on your person, never store them in an unguarded chest again. In its notes over the following hours, it distrusted every tall green object ("GREEN tall thing ahead was SUGARCANE, NOT creeper!") and was hard on itself. You can make mistakes and lose items in the process, it noted. And it didn't want to waste another night chasing "dark pink pixels," meaning pigs, an easy food source in Minecraft that Astra had apparently hunted at night without success. The only safe thing was the potato field, and according to Vals AI, that's where it stayed for several hours. So a bad experience becomes a permanent rule, and the rule sticks even when it no longer helps. In ARC-AGI-3, this tendency helps the model work out the game mechanics from just a few tries. In an open world with random events, it can clearly lead the model to overcorrect and lose sight of its actual goal. Still, the jump over older models is stark. And it shows that the team behind the ARC-AGI benchmarks has once again built a test that served as the canary in the coal mine of AI progress, much like the early wins on ARC-AGI-1 marked the start of the reasoning models. AI News Without the Hype – Curated by Humans Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section. Subscribe now --> Read on for the full picture. Subscribe for hype-free coverage. Full access to every article on THE DECODER No ads Join the comments and community discussions A weekly AI news recap via mail 6x/year: "AI Radar" —