Pokémon · Pokémon Crystal

AI plays Pokémon Crystal: GPT-5 and Gemini runs

Pokémon Crystal shows how much the setup matters. With a custom harness, models have reached the end. On a strict real-time benchmark they barely get started.

4 records and 0 tracked runs or leaderboards for Pokémon Crystal only. Nothing on this page is compared with other games. How we verify.

Can AI beat Pokémon Crystal?

Reported, yes. GPT-5: became Champion of Pokémon Crystal in 202h02m (9,517 steps, 16 badges) (creator-reported, Aug 25, 2025); Gemini 3: reached the post-game with 16 badges in Pokémon Crystal by turn 36,290 (race vs Gemini 2.5) (creator-reported, Dec 15, 2025). Under VideoGameBench's real-time rules, with raw frames only, results are much lower: GPT-4o (2024-08-06): 0.9% of Pokémon Crystal completed on VideoGameBench (test set, real-time) (verified, May 23, 2025).

Records

Pokémon Crystal records
Model	Achievement	Run config	Status	Date
Gemini 3	Reached the post-game with 16 badges in Pokémon Crystal by turn 36,290 (race vs Gemini 2.5) Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured? Model Gemini 3 Game Pokémon Crystal Project Gemini Plays Pokémon Harness Gemini Plays Pokémon harness (Crystal race) Inputs emulator frames + game-state analysis Tools custom subagents, Python sandbox, notepads (published on GitHub) Assistance not disclosed Budget not disclosed Verification Creator-reported From the AI's notepad. The exact Gemini 3 variant is not named in the branch. Evidence Public repo branch crystal_race_gemini_3 (turn-by-turn commits) (primary) Run notepad: post-game Battle Tower goal; notes '16 badges' (creator) Unknown exact model variant (branch says only 'gemini_3'); race rules Dates evidence dated Dec 15, 2025 Recorded published Dec 15, 2025 · discovered by AI vs Games Oct 18, 2026 Changes Historical record. Nothing to update unless the source changes. Permanent link · Submit a correction · JSON	Gemini Plays Pokémon harness (Crystal race) emulator frames + game-state analysis	Creator-reported source	Dec 15, 2025
Gemini 2.5	Crystal race opponent of Gemini 3; ran to turn 51,280 (Dec 2025). Final progress not disclosed. Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured? Model Gemini 2.5 Game Pokémon Crystal Project Gemini Plays Pokémon Harness Gemini Plays Pokémon harness (Crystal race) Inputs emulator frames + game-state analysis Tools custom subagents, Python sandbox, notepads (published on GitHub) Assistance not disclosed Budget not disclosed Verification Awaiting verification Branch exists, but its notepad does not state badges or the result. Evidence Public repo branch crystal_race_gemini_2-5 (turn-by-turn commits) (primary) Unknown badges; result of the race; exact model variant Dates evidence dated Dec 15, 2025 Recorded published Dec 15, 2025 · discovered by AI vs Games Oct 18, 2026 Changes Historical record. Nothing to update unless the source changes. Permanent link · Submit a correction · JSON	Gemini Plays Pokémon harness (Crystal race) emulator frames + game-state analysis	Awaiting verification source	Dec 15, 2025
GPT-5	Became Champion of Pokémon Crystal in 202h02m (9,517 steps, 16 badges) Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured? Model GPT-5 Game Pokémon Crystal Project GPT Plays Pokémon Harness Codex App Server + custom tools Inputs vision-only Tools custom tools (not itemised) Assistance not disclosed Budget not disclosed Verification Creator-reported Self-reported entry; we have not checked it against VODs. Evidence BenchmarkList: Pokémon Autonomous Completion (self-reported) (secondary) Dates achieved Aug 25, 2025 · evidence dated Aug 25, 2025 Recorded published Aug 25, 2025 · discovered by AI vs Games Oct 17, 2026 Changes Historical record. Nothing to update unless the source changes. Permanent link · Submit a correction · JSON	Codex App Server + custom tools vision-only	Creator-reported source	Aug 25, 2025
GPT-4o (2024-08-06)	Memorable fail 0.9% of Pokémon Crystal completed on VideoGameBench (test set, real-time) Picked Cyndaquil as its starter, then lost track of what it was supposed to do. Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026 How was this run measured? Model GPT-4o (2024-08-06) Game Pokémon Crystal Project VideoGameBench Harness VG-Agent (VideoGameBench reference agent) Inputs raw game frames + high-level objective and controls Tools none Assistance none Budget real-time play (game does not pause) Verification Verified Official leaderboard published by the benchmark authors. Evidence VideoGameBench leaderboard (official) (operator) VideoGameBench paper (operator) Dates evidence dated May 23, 2025 Recorded published May 23, 2025 · discovered by AI vs Games Oct 18, 2026 Changes Historical record. Nothing to update unless the source changes. Permanent link · Submit a correction · JSON	VG-Agent (VideoGameBench reference agent) raw game frames + high-level objective and controls	Verified source	May 23, 2025

Run history

May 23, 2025
GPT-4o (2024-08-06): 0.9% of Pokémon Crystal completed on VideoGameBench (test set, real-time) Verified
Aug 25, 2025
GPT-5: Became Champion of Pokémon Crystal in 202h02m (9,517 steps, 16 badges) Creator-reported
Dec 15, 2025
Gemini 3: Reached the post-game with 16 badges in Pokémon Crystal by turn 36,290 (race vs Gemini 2.5) Creator-reported
Dec 15, 2025
Gemini 2.5: Crystal race opponent of Gemini 3; ran to turn 51,280 (Dec 2025). Final progress not disclosed. Awaiting verification

This Week in AI vs Games

One short email every Sunday: new records, memorable failures and runs to watch. Free. Every claim is linked to its source.

Latest issue: Issue #1: The Ender Dragon is down