Ollama Cloud Quota: DeepSeek V3 Burn Rate

DeepSeek V3 (Pro) just wiped out my entire 5-hour Ollama Cloud quota in one session, which makes zero sense if you assume quota is tied to model "intelligence" or size. I expected a cheaper burn rate for a more efficient model, but the reality is completely different.

I ran a benchmark across the available models to figure out exactly how this quota is being calculated. The results prove there is no direct correlation between the model's capability and how fast it eats through your credits. Some "smaller" or supposedly more efficient models are draining the quota just as fast—if not faster—than the heavy hitters.

For anyone trying to optimize their AI workflow or avoid hitting the ceiling mid-project, you can't rely on the model name to guess your remaining time.

If you want to replicate the test or see the specific drain rates, the evaluation script is available here:

https://github.com/Microck/ollama-quota-bench

And the full breakdown of the findings is detailed here:

https://blog.micr.dev/blog/i-made-my-first-benchmark