Google vs Amazon

Google's open source portfolio (44% of their 34 models) allows them to offer 9 free models and achieve the lowest output pricing at $0.040 per 1M tokens, creating a clear developer acquisition funnel. Amazon's closed approach with 5 proprietary models forces a minimum $0.140 per 1M tokens entry point, but delivers consistent function calling across 100% of models versus Google's 47% coverage.

While Google's top model scores 11% higher, Amazon's focused 5-model portfolio achieves nearly identical average performance (43 vs 45) with 85% fewer models. Amazon's Nova 2 Lite at 54/100 actually outperforms 26 of Google's 34 models, suggesting Google's extensive catalog includes significant performance variance and legacy models dragging down their average.

Amazon guarantees function calling on all 5 models (100% coverage) compared to Google's 47% (16/34 models), making Amazon ideal for production applications requiring consistent tool use. Additionally, 80% of Amazon's models support vision capabilities versus Google's 79%, but with a curated selection that eliminates the need to test across 34 options.

Google offers reasoning on 47% of models (16/34) versus Amazon's 20% (1/5), but this masks a strategic difference: Google spreads advanced capabilities across multiple price tiers while Amazon concentrates reasoning in Nova 2 Lite. Google's approach provides 16 reasoning options from free to $12.00 per 1M tokens, while Amazon's single reasoning model simplifies selection but limits pricing flexibility.

Google's portfolio requires evaluating 34 models with prices ranging 300x ($0.040 to $12.00 per 1M tokens) and inconsistent capabilities like function calling (16/34 models). Amazon's streamlined approach offers predictable pricing ($0.140 to $12.50), universal function calling, and 80% vision coverage, reducing evaluation overhead by 85% while maintaining competitive average scores (43 vs 45).