thank you, the model you mention, SQLCoder is fine-tuned for a specific task and would obviously do better on that specific (SQL) task. if that same task were fined-tuned from the two models you compare, I would be thrilled and will willing to bet that fine-tuning from GPT4 will win bigly.

You're comparing two 'world' models with vastly different capabilities. very unfair and probably bordering on misinformation.