It seems to use less than half the tokens for the same task compared to sol, and in some benchmarks closer to 2/3 less tokens. So the actual cost may be roughly the same or cheaper overall.
neuralese is pretty token efficient i guess.
neuralese is pretty token efficient i guess.