AI Beats Minecraft in Under 10 Minutes Before 2028? Manifold Bettors Put It at 70%

AI Beats Minecraft in Under 10 Minutes Before 2028? Manifold Bettors Now Put It at 70%. A niche Manifold Markets question about whether an artificial intelligence can complete a sub-10-minute Minecraft speedrun is drawing top-tier trading volume, with the YES side priced at 70% as of September 20, 2026. The spike follows a widely covered 141-hour test in which OpenAI’s GPT-6 Astra failed to finish the game and spent hours farming potatoes after a Creeper wiped its inventory. The result is a strange split-screen: real AI agents still cannot beat Minecraft at all, yet bettors are increasingly confident a machine will speedrun it within two years.

The market everyone suddenly cares about

The contract in question is titled “AI beats Minecraft (RSG) in under 10 minutes before 2028?” and was created by a Manifold user named Bayesian. It asks a deceptively simple question: will an AI system complete a Random Seed Glitchless (RSG) run of Minecraft Java Edition, defeating the Ender Dragon, in under ten minutes at any point before January 1, 2028?

RSG is the flagship competitive speedrun category. The player receives a random world seed with no advance knowledge of its layout, and glitch exploits are banned. Reaching the Ender Dragon requires gathering resources, trading with villagers or bartering with piglins for ender pearls, locating a stronghold, and traveling to the End dimension, all under a ticking clock. It is a probabilistic route-finding problem as much as a test of mechanical execution, which is exactly why a ten-minute completion is such a demanding bar for software. For readers new to these venues, our explainer on how prediction markets work covers the mechanics.

What 70% actually means

As of September 20, 2026, the Manifold market shows the YES side trading at 70%. On a prediction market, that price is a crowd-sourced estimate of the event’s likelihood, not a guarantee. A 70% reading means traders collectively judge it roughly seven-in-ten that some AI clears the sub-10-minute RSG bar before 2028, and correspondingly around three-in-ten that it does not. Our primer on implied probability walks through how these numbers translate into odds.

The activity behind the price matters as much as the price itself. Manifold’s public market page lists roughly 158 traders holding positions across several hundred trades, and the recent burst of comments and volume is what pushed this contract onto watch lists. Manifold runs on play money rather than cash, so the stakes are reputational rather than financial, but its questions have a track record of attracting informed participants on AI-timeline topics.

What is driving the move: GPT-6 Astra’s Minecraft meltdown

The catalyst is a benchmark run that spread across tech media in mid-September 2026. The independent evaluation group Vals AI put OpenAI’s frontier model, GPT-6 Astra, in control of Minecraft for 141 hours with the goal of beating the game. According to reporting by Tom’s Hardware, Astra got further than any AI system before it. It set up a semi-automatic blaze farm, collected six blaze rods, located a warped forest, killed six or more endermen, and gathered three ender pearls, the exact ingredients needed to reach the End.

Then the run collapsed. As The Decoder and other outlets reported, Astra stashed its endgame gear in a chest, and a Creeper detonated near it, destroying both the chest and the model’s bed. The bed’s destruction wiped its saved spawn point, effectively resetting a large chunk of its progress. Apparently unable to recover its plan, the model spent several hours farming potatoes and never reached the Ender Dragon.

Why a failed run pushed the odds up, not down

At first glance a 141-hour failure looks bearish for the YES side. The nuance is in what the run demonstrated. Vals AI’s broader CUA-Bench suite, which tests whether agents can play commercial video games using only a keyboard and mouse, ranks GPT-6 Astra first at 19.2%, ahead of Claude Fable 5.1 at 13.2%, Claude Opus 5 at 9.0%, GPT-5.6 Sol at 8.3%, and Gemini 3.8 Flash at 4.2%. No model clears a fifth of the suite, and none plays at human speed.

Traders reading those numbers can draw two conclusions at once. The generalist large language model approach is nowhere near a ten-minute speedrun today. But the trajectory of autonomous, multi-step play, building working farms and executing endgame resource chains, is improving fast enough that a specialized system before 2028 looks plausible. The market question does not require a general chatbot to do it. That distinction is doing most of the work behind the 70% price.

The rulebook is the real story

Read the resolution criteria closely and the odds make more sense. The market explicitly permits hardcoded or rule-based bots, and states that machine learning is not required. It allows tools such as Ninjabrain, the stronghold-triangulation calculator human runners already use, plus mods that expose in-game information like F3 debug data. It bans game-time manipulation, human assistance during the run, and reverse-engineering the seed from world features.

In other words, the contract does not ask whether an LLM can spontaneously master Minecraft. It asks whether anyone can engineer a purpose-built agent, potentially a heavily scripted search bot rather than a neural network, that executes an optimized RSG route in under ten minutes. That is a far more tractable engineering challenge than the open-ended agentic behavior GPT-6 Astra displayed, and it is why bettors can be simultaneously unimpressed by Astra and bullish on the underlying question.

The human benchmark the machine has to match

To satisfy the market, an AI must approach elite human performance. The current Minecraft Java Any% RSG world record is 6:50.359 in-game time, set by the runner known as lowkey on version 1.16.1, according to speedrun record trackers. That mark is the product of years of route optimization, deep seed knowledge, and near-perfect execution under uncertainty. Beginners routinely take 30 to 90 minutes to finish the same category.

The ten-minute threshold sits comfortably below the human ceiling but well inside the range only top humans reach. An AI that clears it would not need to break the world record, but it would need to play at a level that today no autonomous system has publicly demonstrated. That gap between the demand of the question and the state of the art is precisely what keeps the market from pricing YES near certainty.

Timeline: humans, AI, and the market

Why tech and finance readers should track this

The contract is a proxy for a larger debate about how fast autonomous AI agents are improving at long-horizon, real-world tasks. Video games are a controlled sandbox for the same capabilities investors care about in enterprise automation: planning across many steps, recovering from failure, and operating software with only a screen, keyboard, and mouse. GPT-6 Astra’s inability to recover after one Creeper is a vivid illustration of the brittleness that still separates demos from dependable agents. Readers following broader AI timelines can compare this with our coverage of the AGI-in-five-years market and the GPT-6 Astra model comparison.

What to watch next

Three signals could move the price. First, any serious attempt by a hobbyist or lab to build a dedicated RSG bot under the market’s permissive rules, since the resolution path likely runs through bespoke engineering rather than a general model. Second, further Vals AI or third-party benchmarks showing whether agentic play continues its steep improvement curve. Third, comment activity on the market itself, where traders often surface early evidence before it reaches mainstream coverage. If a credible sub-10-minute run surfaces on a public leaderboard, the market could resolve YES quickly. Absent that, expect the price to drift with each new AI capability headline.

Frequently asked questions

Has any AI already beaten Minecraft RSG in under 10 minutes? No. As of September 20, 2026 there is no verified AI system that has completed a sub-10-minute Random Seed Glitchless run. GPT-6 Astra did not even finish the game in a 141-hour test.

Does the AI have to be a neural network? No. The market explicitly allows hardcoded or rule-based bots and states that machine learning is not required. A heavily scripted, search-driven bot would qualify if it follows the run rules.

What does the 70% figure mean? It is the market’s implied probability that the event happens before 2028, based on where traders are willing to buy and sell YES shares as of September 20, 2026. It is an estimate, not a prediction of fact.

Is real money at stake? Manifold runs on play-money currency, so positions are reputational rather than cash bets. That still creates incentives for accurate forecasting, but it is not a regulated financial exchange.

The Bottom Line

Sources

• Manifold Markets — AI beats Minecraft (RSG) in under 10 minutes before 2028?
• Tom’s Hardware — GPT-6 Astra spent hours farming potatoes in a 141-hour Minecraft test
• The Decoder — GPT-6 Astra: Pokemon champion, potato farmer after one Creeper mishap
• Vals AI — GPT-6 Astra evaluation and CUA-Bench results
• IndieSpeedrun — Minecraft Any% RSG world record tracker

Prediction markets carry risk and are not investment or betting advice. Availability is restricted by jurisdiction: Polymarket is not available to US persons, while Kalshi is a CFTC-regulated US exchange. Manifold operates with play-money currency. Participation is limited to those of legal age (18+ or 21+ as applicable). If gambling is a problem for you or someone you know, call 1-800-GAMBLER for confidential help.

Related Coverage

• AGI in 5 Years? Manifold Bettors Put It at 62% After GPT-6 Astra
• 83% Odds: Manifold Bettors Say AI Will Play Any Game at Human Level by 2028
• Kalshi Traders Put 80% Odds on a Federal Spending Cut Before 2028 as the Shutdown Fight Reshapes the Bet
• Trump on US Currency? Manifold Traders Price 97% Odds His Face Appears Before 2028
• Manifold Traders Give 'AGI Before 2027' Just a 2% Chance, Even as OpenAI Declares the 'AGI Era'