How fast are Cerebras models?

Here are the output speeds (tokens per second) of each of our available models:

• llama3.1-8b: ~2200 tok/sec

• llama-3.3-70b: ~2100 tok/sec

• llama-4-scout-17b-16e-instruct: ~2600 tok/sec

• qwen-3-32b: ~2100 tok/sec

• deepseek-r1-distill-llama-70b: ~1700 tok/sec