Sakana - Fugu Ultra did not do well in my AI game tech demo's.
This is a model being talked about as Fable/Mythos level, but my marble puzzle benchmark was honestly disappointing.
I’ve run the same prompt across multiple models, and this was the worst result of the bunch.
It also cost me about $20 on pay-as-you-go, which made it more expensive than GLM 5.2 for this test.
I didn't like the following, when comparing to the model tests :
-The marble physics didn’t feel or behave real.
-The wooden textures were low quality.
-The maze/path didn’t look properly constructed.
And overall, the result lacked the polish and coherence
I’d expect from a model being compared to S tier frontier-level systems.
I’ll test my other AI game tech demos with it if people want, but based on this first run, I came away disappointed.
00:34