August 2026 Summaries

A benchmark conducted over 5–6 August 2026 compared six AI gateways—Edgee, Cloudflare, TrueFoundry, Vercel, OpenRouter, and Kong—with direct OpenAI and Anthropic calls, measuring time to first token across 16,344 requests from 43 Google Cloud regions. For GPT-5.4, Edgee recorded the lowest gateway median latency at 798 ms, or 24 ms above direct OpenAI access, while Cloudflare led on Claude Sonnet 4.6 at 1,140 ms, with Edgee second at 1,215 ms. Run-by-run comparisons across 11,600 matched tests found Edgee returned the first token before competitors in 66.7% of cases, although its results against Cloudflare were effectively tied. The analysis reported substantially higher overhead for OpenRouter and especially Kong, with Kong appearing not to stream Anthropic responses, which contributed to its much longer first-token times. The author notes that the limited two-day test, short prompts, and two-model scope cannot establish performance for long contexts or tool calls, and has made the Node.js-based benchmark publicly available for independent review and replication.