Consider changing model, verify the following: - Parallelization (with rate-limit validation, possibly using multiple API keys) - Token pricing - Performance / speed - Benchmarks We want tool to work sensibly with the worst of currently popular models.