The new run has started. Tune in! twitch.tv/gemini_plays_p…

yoho..!🐟

My Gemini Plays Pokémon project got a shoutout from Sundar Pichai at #GoogleIO 🤘🔥

ClaudePlaysPokemon restarted with Claude 4 so for fun we're restarting too! You'll be able to watch Claude and Gemini play side-by-side, exploring each model and their harnesses' strengths and weaknesses! (Note: don't treat this as a serious race!) holodex.net/multiview/AAGY…

Watch o3 play Pokémon—live. See how it plans its next move, explains its reasoning, analyzes the map visually, and saves to memory. Thank you community member Clad3815 for putting this stream together!

Test run of Pkmn Yellow Legacy is live! This new run comes with an updated harness: ➕notepad ➖predefined agents deleted ➕AI creates custom agents ➕code execution ... and more! Notepad and agents are tracked live: github.com/waylaidwandere… Watch now: twitch.tv/gemini_plays_p…

this thing in the Gemini 2.5 tech report actually redeems Gemini Plays Pokemon quite a bit - a lot of people were complaining that Gemini Plays Pokemon started with a Blastoise and received constant developer edits during the run (and therefore is not comparable with Claude

I'm working on a post about my Gemini Plays Pokemon project. This will be the first developer blog post I've ever written. I hope to give you some interesting behind-the-scenes information. Follow to be notified: blog.jcz.dev

It's my birthday today! As a gift, please come by the stream and say hi to Gem 🥳 Still streaming Pokemon Yellow Legacy (hard mode), and Gemini is making good progress with its new harness that lets it code and create its own agents freely! twitch.tv/gemini_plays_p…

Short test stream of Gemini Plays Pokemon Crystal starting now on channel 2: twitch.tv/genai_plays

Beautiful Suisei keyboard 🙏

This shows how even a simple notepad can make a big difference for long-horizon tasks. In the current Pokémon Yellow Legacy run, Gemini uses one to keep track of goals, plans, and issues. twitch.tv/geminiplayspok…

This is on my roadmap for Gemini Plays Pokemon 🔥

As models improve, complex scaffolding becomes less necessary — a strong system prompt is often enough. Benchmarks like Gemini Plays Pokémon help show where today’s SOTA models fall along that spectrum.

I wrote up the making-of for Gemini Plays Pokémon: how I designed the scaffold so Gemini 2.5 Pro could handle a long-horizon game, what failed, and the lessons that made it work. Full post: blog.jcz.dev/the-making-of-…

Found #IRyStocrats in the wild today #IRyS

Still one of the coolest Pokemon cards I own! x.com/TheCodeOfJoel/… #galaxillust

🎮✨ Gemini 2.5 Pro took on Pokémon Blue and became the champion. Twice. Now it’s tackling even tougher challenges in Pokémon Yellow Legacy, and is creating its own tools and subagents as it goes. Tune in to the Elite Four live now on hard mode with level caps, no items in

Andrew Carr (e/🤸) hey heads up, this is inaccurate on multiple levels -- Gemini 2.5 Pro finished in 35k actions (see arxiv.org/pdf/2507.06261 Fig 15a), Claude has not finished at all, the definition of a "step" is different across all three (see the 2.5 report), and tools/harnesses are different

I've been seeing a few posts lately comparing GPT-5 vs o3 on Pokemon that have some misconceptions - to exemplify the problem with these comparisons, one "step" for GPT-5 can include a lookup to game knowledge that instantly solves certain complex puzzles like Cinnabar Mansion