They have cache, but it costs the same indeed, no idea what the point of the cache is
Say avg model response length is 1024 tok. At 50 tok/s normal providers do your turn will only take 20s (vs Cerebras 101s) and will cost 20x less. That time and cost is per single tool call.