| Gemini 3.8 Flash |
Memorable fail 3 of 8 badges in Pokémon Blue, run finished after 5,648 turns Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.8 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness (mGBA emulator)
- Inputs
- emulator frames + game-state analysis
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Verified Hub game-state telemetry. The game is inferred from the hub run name ('blue_'); this run has no public repo branch.
- Evidence
- Hub telemetry: run blue_gemini_3_8_flash (primary)
- Unknown
- why the run was marked finished; whether it was a timed or capped run
- Dates
- started Oct 6, 2026 · evidence dated Oct 14, 2026
- Recorded
- published Oct 14, 2026 · discovered by AI vs Games Oct 18, 2026
- Changes
- No change detected since we started tracking.
Permanent link · Submit a correction · JSON |
Gemini Plays Pokémon harness (mGBA emulator) emulator frames + game-state analysis |
Verified |
Oct 14, 2026 |
| Gemini 4 Argon |
8 of 8 badges in Pokémon Blue, run finished after 1,753 turns Same game, same hub: Gemini 4 Argon reached 8 badges in about a third of the turns Gemini 3.8 Flash used for 3. Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 4 Argon
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness (mGBA emulator)
- Inputs
- emulator frames + game-state analysis
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Verified Hub game-state telemetry. The game is inferred from the hub run name ('blue_'); this run has no public repo branch.
- Evidence
- Hub telemetry: run blue_gemini_4_argon (primary)
- Unknown
- why the run was marked finished; whether it was a timed or capped run
- Dates
- started Oct 10, 2026 · evidence dated Oct 11, 2026
- Recorded
- published Oct 11, 2026 · discovered by AI vs Games Oct 18, 2026
- Changes
- No change detected since we started tracking.
Permanent link · Submit a correction · JSON |
Gemini Plays Pokémon harness (mGBA emulator) emulator frames + game-state analysis |
Verified |
Oct 11, 2026 |
| Gemini 3.8 Flash |
8 of 8 badges in Pokémon Blue (47,889 turns, idle) Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.8 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-first' setup
- Inputs
- vision-first (exact inputs not documented)
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Verified Badges from hub telemetry; game confirmed by the run's notepad in the public repo.
- Evidence
- Hub telemetry: run vision_first_gemini_3_8_flash (primary)
Public repo branch vision_first_gemini_3_8_flash (turn-by-turn commits) (primary) Run notepad (names the game as Pokémon Blue) (creator)
- Dates
- started Sep 3, 2026 · evidence dated Sep 27, 2026
- Recorded
- published Sep 27, 2026 · discovered by AI vs Games Oct 18, 2026
- Changes
- No change detected since we started tracking.
Permanent link · Submit a correction · JSON |
Gemini Plays Pokémon harness, 'vision-first' setup vision-first (exact inputs not documented) |
Verified |
Sep 27, 2026 |
| Gemini 3.8 Flash |
Became Champion of Pokémon Blue at turn 24,981 (Hall of Fame at turn 24,994) Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.8 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-first' setup
- Inputs
- vision-first (exact inputs not documented)
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Creator-reported Turn numbers come from the AI's own notepad checklist, not from game state.
- Evidence
- Hub telemetry: run vision_first_gemini_3_8_flash (primary)
Public repo branch vision_first_gemini_3_8_flash (turn-by-turn commits) (primary) Run notepad: Champion at turn 24,981, Hall of Fame at turn 24,994 (creator)
- Dates
- started Sep 3, 2026 · evidence dated Sep 27, 2026
- Recorded
- published Sep 27, 2026 · discovered by AI vs Games Oct 18, 2026
- Changes
- Historical record. Nothing to update unless the source changes.
Permanent link · Submit a correction · Share card (PNG) · JSON |
Gemini Plays Pokémon harness, 'vision-first' setup vision-first (exact inputs not documented) |
Creator-reported |
Sep 27, 2026 |
| Gemini 3.7 Flash |
8 of 8 badges in Pokémon Blue (55,601 turns, idle) Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.7 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-first' setup
- Inputs
- vision-first (exact inputs not documented)
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Verified Badges from hub telemetry; game confirmed by the run's notepad in the public repo.
- Evidence
- Hub telemetry: run vision_first_gemini_3_7_flash (primary)
Public repo branch vision_first_gemini_3_7_flash (turn-by-turn commits) (primary) Run notepad (names the game as Pokémon Blue) (creator)
- Dates
- started Aug 16, 2026 · evidence dated Sep 19, 2026
- Recorded
- published Sep 19, 2026 · discovered by AI vs Games Oct 17, 2026
- Changes
- No change detected since we started tracking.
Permanent link · Submit a correction · JSON |
Gemini Plays Pokémon harness, 'vision-first' setup vision-first (exact inputs not documented) |
Verified |
Sep 19, 2026 |
| Gemini 3.7 Flash |
Became Champion of Pokémon Blue at turn 39,310 Gemini 3.8 Flash got there about 14,000 turns sooner on the same vision-first setup. Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.7 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-first' setup
- Inputs
- vision-first (exact inputs not documented)
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Creator-reported Turn number comes from the AI's own notepad checklist, not from game state.
- Evidence
- Hub telemetry: run vision_first_gemini_3_7_flash (primary)
Public repo branch vision_first_gemini_3_7_flash (turn-by-turn commits) (primary) Run notepad: 'Become Pokémon League Champion and Enter Hall of Fame' at turn 39,310 (creator)
- Dates
- started Aug 16, 2026 · evidence dated Sep 19, 2026
- Recorded
- published Sep 19, 2026 · 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, 'vision-first' setup vision-first (exact inputs not documented) |
Creator-reported |
Sep 19, 2026 |
| Gemini 3.5 Flash |
Memorable fail 6 of 8 badges in Pokémon Blue (74,636 turns, idle) Stalled on 6 badges after 74,636 turns, the longest run in the series and the only one without all 8. Last verified: Oct 18, 2026 · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.5 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-first' setup
- Inputs
- vision-first (exact inputs not documented)
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Verified Badges from hub telemetry; game confirmed by the run's notepad in the public repo.
- Evidence
- Hub telemetry: run vision_first_gemini_3_5_flash (primary)
Public repo branch vision_first_gemini_3_5_flash (turn-by-turn commits) (primary) Run notepad (names the game as Pokémon Blue) (creator)
- Dates
- started Aug 16, 2026 · evidence dated Sep 3, 2026
- Recorded
- published Sep 3, 2026 · discovered by AI vs Games Oct 17, 2026
- Changes
- No change detected since we started tracking.
Permanent link · Submit a correction · JSON |
Gemini Plays Pokémon harness, 'vision-first' setup vision-first (exact inputs not documented) |
Verified |
Sep 3, 2026 |
| Gemini 3.6 Flash |
Memorable fail 4 of 8 badges in Pokémon Blue at turn ~55,800 (last repo commit Aug 15, 2026) Spent more than 55,000 turns and still had only 4 badges, stuck around Lavender Town. Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3.6 Flash
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-first' setup
- Inputs
- vision-first (exact inputs not documented)
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Creator-reported Not listed on the hub; badge count is from the AI's notepad.
- Evidence
- Public repo branch vision_first_gemini_3_6_flash (turn-by-turn commits) (primary)
Run notepad: 'Badges (4/8)' at turn 55,771 (creator)
- Dates
- evidence dated Aug 15, 2026
- Recorded
- published Aug 15, 2026 · 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, 'vision-first' setup vision-first (exact inputs not documented) |
Creator-reported |
Aug 15, 2026 |
| Gemini 3 Pro |
All 8 badges in a vision-only Pokémon Blue run (last turn 67,315, May 2026) Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 3 Pro
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness, 'vision-only' setup (v2)
- Inputs
- vision-only
- Tools
- custom subagents, Python sandbox, notepads (published on GitHub)
- Assistance
- not disclosed
- Budget
- not disclosed
- Verification
- Creator-reported Badge list is from the AI's notepad; this run is not on the hub.
- Evidence
- Public repo branch gemini_3_pro_vision_only_blue_v2 (turn-by-turn commits) (primary)
Run notepad lists all 8 badges (creator)
- Unknown
- whether the run reached Champion
- Dates
- evidence dated May 23, 2026
- Recorded
- published May 23, 2026 · 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, 'vision-only' setup (v2) vision-only |
Creator-reported |
May 23, 2026 |
| Gemini 2.5 Pro |
Gemini 2.5 Pro beats Pokémon Blue with no manual interventions (~406 hours, 36,801 actions) Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 2.5 Pro
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness (2025): Node.js + Lua bridge to mGBA; RAM data plus annotated screenshots; critique, summary and goal prompts; map memory; Pathfinder and Boulder Puzzle Strategist agents (finalized)
- Inputs
- RAM data + annotated screenshots
- Tools
- Pathfinder agent, Boulder Puzzle Strategist agent, map memory, critique/summary/goals prompts
- Assistance
- none (creator-reported)
- Budget
- not disclosed
- Verification
- Creator-reported Stated by the run's creator; not independently checked by us.
- Evidence
- Creator's write-up: The Making of Gemini Plays Pokémon (creator)
- Unknown
- exact ROM version; reasoning budget; public VOD or log for this run
- Dates
- achieved Jun 9, 2025 · evidence dated Jun 9, 2025
- Recorded
- published Jun 9, 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 (2025): Node.js + Lua bridge to mGBA; RAM data plus annotated screenshots; critique, summary and goal prompts; map memory; Pathfinder and Boulder Puzzle Strategist agents (finalized) RAM data + annotated screenshots |
Creator-reported |
Jun 9, 2025 |
| Gemini 2.5 Pro |
Gemini 2.5 Pro becomes Champion in Pokémon Blue (development run, ~813 hours) Last verified: not independently verified · Last checked: Oct 18, 2026 How was this run measured?
- Model
- Gemini 2.5 Pro
- Game
- Pokémon Blue
- Project
- Gemini Plays Pokémon
- Harness
- Gemini Plays Pokémon harness (2025): Node.js + Lua bridge to mGBA; RAM data plus annotated screenshots; critique, summary and goal prompts; map memory; Pathfinder and Boulder Puzzle Strategist agents
- Inputs
- RAM data + annotated screenshots
- Tools
- Pathfinder agent, Boulder Puzzle Strategist agent, map memory, critique/summary/goals prompts
- Assistance
- manual interventions and harness changes during the run (development run)
- Budget
- not disclosed
- Verification
- Creator-reported Stated by the run's creator; not independently checked by us.
- Evidence
- Creator's write-up: The Making of Gemini Plays Pokémon (creator)
Google CEO on X: 'Gemini 2.5 Pro just completed Pokémon Blue!' (May 3, 2025 UTC) (secondary) TechCrunch: Google's Gemini has beaten Pokémon Blue (with a little help) (secondary)
- Unknown
- exact ROM version; step or action count for this run; reasoning budget; full list of manual interventions
- Dates
- achieved May 2, 2025 · evidence dated May 2, 2025
- Recorded
- published May 2, 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 (2025): Node.js + Lua bridge to mGBA; RAM data plus annotated screenshots; critique, summary and goal prompts; map memory; Pathfinder and Boulder Puzzle Strategist agents RAM data + annotated screenshots |
Creator-reported |
May 2, 2025 |