OpenAI Models Speedtest

A client-side streaming benchmark for comparing time to first token, end-to-end latency, and output throughput. Your API key is used only for direct requests to api.openai.com.

Benchmark setup

OpenAI API key

Saved in this browser's localStorage. It never leaves the browser except in the Authorization header sent to api.openai.com.

Tries per model × modeTotal budget cap (USD)Fixed prompt

Ready. Add models, prices, then run.

Models and editable prices

Prices are editable estimates and can drift; verify them against OpenAI's current pricing page before using this for cost decisions. “FAST” sends service_tier: "priority"; unsupported models are recorded as failures without substituting another model.

Test?	Model	Mode	Normal input $/1M	Normal output $/1M	FAST input $/1M	FAST output $/1M

Cost is computed from returned input_tokens and output_tokens; cached/reasoning tokens are not separately priced by this lightweight tool.

Results

$0.0000 / $0.000 / 0 tries

Results will appear as attempts finish.

Model	Mode	Done	Failures	TTFT median (min–max)	Total median	Tokens/sec median	Cost	Throughput	TTFT (lower is better)
No results yet.

Measurements use the Responses API streaming endpoint. TTFT starts immediately before fetch() and ends on the first response.output_text.delta, so it includes browser/network/API queue time.