PolarGrid vs Together AI vs Baseten vs Fireworks vs Anthropic vs OpenAI

Same prompt, measured client-side from your browser. Every provider is called through the same server-side proxy, so no lane gets a network head start. PolarGrid serves Qwen 3.5 27B at the edge; Groq serves Qwen3.6 27B (the closest same-size comparison); Together AI, Baseten and Fireworks serve open-weight gpt-oss models (20B and 120B) from centralized cloud, so those lanes compare serving infrastructure across different open models. Anthropic and OpenAI run their own models. gpt-oss reasons before answering — reasoning tokens count toward TTFT and tok/s but are not displayed. Throughput is omitted for Anthropic (its stream chunks are not 1:1 tokens).