Hacker News

Gemini 3 Pro vs. 2.5 Pro in Pokemon Crystal

• 43 Comments
• So after years of being gleefully told that AI will replace all jobs an omniscient state of the art model, with heavy assistance, takes more than two weeks and thousands of dollars in tokens to do what child me did in a few days? Huh.

  By sussmannbaka 13 minutes ago

• I used to think the same until latest agents started adding perfectly fine features to a large existing react app with just basic input (in English) . Most of the jobs require levels of intelligence below that. It's just a matter of time before agents get to that.

  By murukesh_s 2 minutes ago

• 1.88 billion tokens * $12 / 1M tokens (output) suggests a total cost of $22,560 to solve the game with Gemini 3 Pro?

  By bbondo 3 hours ago

• “Gemini 3 Pro was often overloaded, which produced long spans of downtime that 2.5 Pro experienced much less often”
  I was unclear if this meant that the API was overloaded or if he was on a subscription plan and had hit his limit for the moment. Although I think that the Gemini plans just use weekly limits, so I guess it must be API.

  By elephanlemon 2 hours ago

• True though I bet the $200 a month plan could do it, maybe a few extra days of downtime when quota was maxed

  By brianwawok 2 hours ago

• For how long would it stay $200 of you can rack up 5 figures if usage..

  By AstroBen an hour ago

• That is the reason they severely limited Claude Max subscriptions. Some users racked up 1k+ in API equivalent cost per day.

  By manmal an hour ago

• I can't believe how massively underpaid I was when I was 11

  By mkoubaa 2 hours ago

• Do you hallucinate as a kid?

  By re-thc 2 hours ago

• At that age, it's called "imagination"

  By foundddit an hour ago

• Kids definitely do this. They fill in blanks/context with assumptions, resulting in all sorts of silly responses, for topics of sparse knowledge/certainty. They're not lying, because they think it's true. Sometimes the gap filling is wrong, but usually downright brilliant, within the context of their knowledge.

  By nomel 14 minutes ago

• :/ Damn. That needs to cost 1000x less before people can try it on their own games.

  By ogogmad 2 hours ago

• That's an extrapolation to finish the entire game.
  If limit your token count to a fraction of 2 billion tokens, you can try it on your own game, and of course have it complete a shorter fraction of the game.

  By someperson an hour ago

• "Crucially, it tells the agent not to rely on its internal training data (which might be hallucinated or refer to a different version of the game) but to ground its knowledge in what it observes. "
  Does this even have any effect?

  By oceansky 3 hours ago

• Yes, at least to some extent. The author mentions that the base model knows the answer to the switch puzzle but does not execute it properly here.
  "It is worth noting that the instruction to "ignore internal knowledge" played a role here. In cases like the shutters puzzle, the model did seem to suppress its training data. I verified this by chatting with the model separately on AI Studio; when asked directly multiple times, it gave the correct solution significantly more often than not. This suggests that the system prompt can indeed mask pre-trained knowledge to facilitate genuine discovery."

  By ragibson 3 hours ago

• My issue with this is that the LLM could just be roleplaying that it doesn't know.

  By hypron 3 hours ago

• Of course it is. It's not capable of actually forgetting or suppressing its training data. It's just double checking rather than assuming because of the prompt. Roleplaying is exactly what it's doing. At any point, it may stop doing that and spit out an answer solely based on training data.
  It's a big part of why search overview summaries are so awful. Many times the answers are not grounded in the material.

  By jdiff 2 hours ago

• To test would just need to edit the rom and switch around the solution. Not sure how complicated that is, likely depends on the rom system.

  By brianwawok 2 hours ago

• I don't know why people still get wrapped around the axle of "training data".
  Basically every benchmark worth it's salt uses bespoke problems purposely tuned to force the models to reason and generalize. It's the whole point of ARC-AGI tests.
  Unsurprisingly Gemini 3 pro performs way better on ARC-AGI than 2.5 pro, and unsurprisingly it did much better in pokemon.
  The benchmarks, by design, indicate you can mix up the switch puzzle pattern and it will still solve it.

  By Workaccount2 2 hours ago

• I'm wondering about this too. Would be nice to see an ablation here, or at least see some analysis on the reasoning traces.
  It definitely doesn't wipe its internal knowledge of Crystal clean (that's not how LLMs work). My guess is that it slightly encourages the model to explore more and second-guess it's likely very-strong Crystal game knowledge but that's about it.

  By tootyskooty 3 hours ago

• The model probably recognizes the need for a grassroots effort to solve the problem, to "show it's work".

  By Workaccount2 2 hours ago

• It will definitely have some effect. Why won't it? Even adding noise into prompts (like saying you will be rewarded $1000 for each correct answer) has some effect.
  Whether the 'effect' something implied by the prompt, or even something we can understand, is a totally different question.

  By raincole 2 hours ago

• Do we have examples of this in promps in other contexts?

  By baby 2 hours ago

• I very much doubt it

  By blibble 3 hours ago

• If they trained the model to respond to that, then it can respond to that, otherwise it can't necessarily.

  By astrange 2 hours ago

• I think you got a point here. These companies are injecting a lot of datasets every day into it.

  By oceansky 2 hours ago

• It might get things wrong on purpose, but deep down it knows what it's doing

  By mkoubaa 2 hours ago

• Nice writeup! I need to start blogging about my antics. I rigged up several cutting edge small local models to an emulator all in-browser and unsuccessfully tried to get them to play different Pokémon games. They just weren't as sharp as the frontier models.
  This was a good while back but I'm sure a lot of people might find the process and code interesting even if it didn't succeed. Might resurrect that project.

  By soulofmischief 3 hours ago

• I have to think they need to know enough of the guides for the game for it to work out, how do they know whats on screen?

  By giancarlostoro 3 hours ago

• In my project I rigged up an in-browser emulator and directly fed captured images of the screen to local multimodal models.
  So it just looks right at what's going on, writes a description for refinement, and uses all of that to create and manage goals, write to a scratchpad and submit input. It's minimal scaffolding because I wanted to see what these raw models are capable of. Kind of a benchmark.

  By soulofmischief 3 hours ago

• I have a feeling if you gave them access to GameFAQ guides they might be able to play better, but it depends on how you can feed them the data.

  By giancarlostoro 5 minutes ago

• I like the inclusion of the graph at the end to compare progress. It would be cool to compare this directly to competing models (Claude, GPT, etc).

  By cg5280 2 hours ago

• It would unfortunately also need several runs of each to be reliable. There's nothing in TFA to indicate the results shown aren't to a large degree affected by random chance!
  (I do think from personal benchmarks that Gemini 3 is better for the reasons stated by the author, but a single run from each is not strong evidence.)

  By kqr an hour ago

• TFA says multiple times that the results are affect by random chance

  By casey2 an hour ago

• Yes, but recognising that is only the first step. Quantifying the variance is the next step which I miss in the article.

  By kqr 41 minutes ago

• How certain can we be that these improvements aren't just a result of Gemini 3 Pro pre-training on endless internet writeups of where 2.5 has struggled (and almost certainly what a human would have done instead)?
  In other words, how much of this improvement is true generalization vs memorization?

  By squimmy26 2 hours ago

• You're too kind. Even the CEO of Google retweeted how well Gemini 2.5 did on Pokemon. There is a high chance that now it's explicitly part of the training regime. We kind of need a different kind of game to know how well it generalizes.

  By zurfer 2 hours ago

• Being through the game recently, I am not surprised Goldenrod Underground was a challenge, it is very confusing and even though I solved it through trial and error, I still don't know what I did. Olivine Lighthouse is the real surprise, as it felt quite obvious to me.

  By jwrallie 3 hours ago

• I wonder how much of it is due to the model being familiar with the game or parts of it, be it due to training of the game itself, or reading/watching walkthroughs online.

  By wild_pointer 3 hours ago

• There was a well-publicised "Claude plays Pokémon" stream where Claude failed to complete Pokemon Blue in spectacular fashion, despite weeks of trying. I think only a very gullible person would assume that future LLMs didn't specifically bake this into their training, as they do for popular benchmarks or for penguins riding a bike.

  By andrepd 3 hours ago

• > as they do for popular benchmarks or for penguins riding a bike.
  Citation?

  By ctoth 15 minutes ago

• While it is true that model makers are increasingly trying to game benchmarks, it's also true that benchmark-chasing is lowering model quality. GPT 5, 5.1 and 5.2 have been nearly universally panned by almost every class of user, despite being a benchmark monster. In fact, the more OpenAI tries to benchmark-max, the worse their models seem to get.

  By criley2 3 hours ago

• Hm? 5.1 Thinking is much better than 4o or o3. Just don't use the instant model.

  By astrange 2 hours ago