Recently, in a global AI showdown, OpenAI's latest model, GPT-5, successfully cleared the classic game "Pokémon Crystal" in a staggering 9,517 steps, defeating its ultimate opponent, Red. This performance not only had livestream viewers shouting "GG" but also pushed AI gaming efficiency to a new level. According to Twitter blogger Clad3815, its predecessor, O3, took 27,040 steps and nearly a month to complete the game, while GPT-5 completed it in just 202 hours.
GPT-5's gameplay was as efficient as the anime character Satoshi Ketchum: collecting 16 badges in just 9,205 steps (O3 took 22,334), and in the final stage, it crushed O3's 5,000 steps with just 312. In the Elite Four Challenge, GPT-5's 7,329 steps compared to O3's 18,000, again demonstrating a tenfold improvement in efficiency. Notably, GPT-5 had already cleared Pokémon Red, establishing an overwhelming advantage. OpenAI co-founder Greg Brockman publicly praised its performance, while technical analyst Clad3815 identified three key factors: reduced hallucinations, enhanced spatial reasoning, and optimized goal planning, enabling it to navigate complex mazes with precision.
The Pokémon game is becoming a "new benchmark" for AI capabilities, but the challenge is costly—users estimate the API cost of GPT-5 completing Pokémon Red at approximately $3,500. This breakthrough not only validates the decision-making and control capabilities of large-scale language models but also provides important insights for the future application of AI in complex tasks.
