AI Stream Studio
RAIS Protocol
RAIS Spec ↗GitHub
Public Benchmarks

See how providers stack up.

Real benchmark runs — latency, throughput, cost, and stability. Every report is reproducible: click any card to view full results or clone the run with your own keys.

⚡ Run your own →Inspect any endpoint
Featured reports
4
Providers covered
5
All reports
URL-encoded · no backend
Protocol
RAIS v1
Groq
OpenAI
Anthropic
May 14, 2026

Groq vs OpenAI vs Anthropic — Speed Shootout

Three cloud providers head-to-head on a short generation task. Groq dominates on raw throughput; Anthropic leads on coherence.

"Explain how transformer attention works in 3 sentences."
⚡ Fastest TTFB
Groq · 248ms
🚀 Best Tok/s
Groq · 207
cloudthroughputcost5 runs
View report →
Local RAIS
Groq
May 15, 2026

Local RAIS vs Groq — On-device vs Cloud

A self-hosted RAIS endpoint against Groq's cloud inference. Local wins on latency when on the same machine; Groq wins on throughput.

"What is the difference between SSE and WebSocket streaming?"
⚡ Fastest TTFB
Local RAIS · 12ms
🚀 Best Tok/s
Groq · 198
localself-hostedprivacy3 runs
View report →
Groq
Ollama
May 13, 2026

Groq vs Ollama — Cloud vs Edge Inference

Groq cloud vs local Ollama (Llama 3.2 3B) on a reasoning task. Cloud wins on speed; Ollama wins on cost and privacy.

"List 5 practical use cases for AI streaming in production apps."
⚡ Fastest TTFB
Ollama · 42ms
🚀 Best Tok/s
Groq · 189
localcloudcost3 runs
View report →
Groq
OpenAI
Anthropic
May 12, 2026

All 3 Cloud Providers — Long Generation

Extended output task across Groq, OpenAI, and Anthropic. Cost-efficiency ratios become visible at higher token counts.

"Write a detailed technical comparison of REST, GraphQL, and SSE for real-time AI…"
⚡ Fastest TTFB
Groq · 261ms
🚀 Best Tok/s
Groq · 214
cloudlong-formcost3 runs
View report →

Submit your benchmark

Run a benchmark on any RAIS-compatible endpoint, generate a share link, and open a PR to add it here. We feature benchmarks that are reproducible, honest, and interesting.

Run a benchmark →Submit via GitHub →
How reports work: Every benchmark is encoded entirely in its URL — no backend, no accounts, no tracking. Anyone with the link can view results, clone the configuration, and re-run with their own API keys. Reports are permanent and portable.