Grok 4.3 on Amazon Bedrock: We Routed Four Frontier Models Through One Gateway and Measured the Cost

TrueFoundry tested four frontier LLMs on Amazon Bedrock using an AI Gateway fallback chain, finding a 30x cost spread ($0.00046 for MiniMax M2.5 to $0.02764 for Claude Opus 4.8) and 9x latency variance on identical prompts. The results show cost and latency pull in opposite directions, arguing for intelligent routing over single-model defaults.