⚡
slowtest.ai

The Speedtest for Artificial Intelligence

How it works How we measure FAQ Insights Top 10 Also at: bogtest.com System Ready

INDEPENDENT · NO ACCOUNT REQUIRED

AI speed test.
Know how fast your AI responds.

Measure tokens per second with a clear clock and a format check.
One benchmark. Three ways to run it.

Best

Send your agent here

Send your agent to the test page. It checks its tools and runs the test if possible.

Agent test page ↗Analyze agent callback ↗
How does it work?

Give your agent this instruction: “Go to https://slowtest.ai/agents and test yourself.” Agents with HTTP POST tools return a server-timed receipt. Read-only agents return audit output through the callback page for analysis. No API key needed.

Good

Your agent submits the test

Copy the test instructions into your AI session.

How does it work?

Copy a test script with a unique run ID. Your agent sends the start signal, writes the audit, then submits the finish signal. Keep this page open to receive the result.

Works

Test and time yourself

Run the prompt, time it, and paste the response.

How does it work?

Copy the prompt into a fresh AI chat. Start the timer when you send it and stop at the last visible word. Paste the response to calculate estimated speed. Manual results stay out of the public high-score board.

How we measure

Top 5 high scores

Model bragging rights. Can your AI beat the board? Take the challenge and share your result receipt. Read the speed-test guides.

Model and harness are self-reported. Generation is the server clock. Full Top 10

Loading verified scores…

    HOW IT WORKS

    Own-key, 1-paste, two clocks, then fidelity

    Slowtest is a clock and a compliance check in your browser. The public high-score board lists eligible server-timed runs, not overall model quality. Manual times are excluded. Each result states its clock; model and harness identities are self-reported.

    Own-key vs 1-paste

    Own-key streams from your browser to a provider or local Ollama. The key stays in localStorage. That path gives a first-to-last Generation window when the stream actually tokens.

    1-paste copies a serial script. The model must POST start, write the audit, then POST finish. This page times Generation from those two server receipts. The result returns to this page automatically.

    Generation tok/s vs full-turn

    Generation tok/s is the decode-ish clock when we have a real generation window (own-key first-to-last token, or 1-paste start/finish beacons).

    Full-turn tok/s is the whole job at the keyboard: prefill, hidden reasoning, tools, queue, copy slack, and the telemetry POST. It is not decode-only. Think time is not Generation.

    Fidelity

    Not IQ. Line 1 must declare MODEL:. The essay must land 800–1200 words on the first try — about 1000 words, one pass, no rewrite. A slow run at 100% fidelity still passes. A fast run that skips Line 1 or overshoots 1200 words is a miss. Speed with a miss is not a score.

    Longer write-up: How we measure · Insights
    Same audit, side by side: Claude Opus 5 vs Grok 4.6 · Claude Fable 5 vs Gemini 3.7 Flash · GPT-5.6 vs DeepSeek V4 · GPT-5.6 vs Grok 4.6 · Claude Opus 5 vs GPT-5.6 · Gemini 3.7 Flash vs Grok 4.6

    FAQ

    AI speed test questions people actually ask

    Honest answers. No invented tok/s bands. If you want a number, run the test.

    What is an AI speed test?

    An AI speed test measures how fast a model produces tokens in a real session — not a vendor slide. Slowtest runs in your browser. You either stream with your own API key or paste a script into a chat. The useful numbers are tokens per second on a stated clock (Generation vs full-turn), plus whether the model followed the audit format (fidelity). No account. Usage stats stay on Vercel.

    What is tokens per second (tok/s)?

    Tokens per second is output throughput: estimated tokens divided by the seconds on a chosen clock. Slowtest estimates tokens from the response text. It does not publish a normal 2026 score and the high-score board uses actual eligible server-timed runs. Run the same ~1000-word audit twice if you want a pair. Empty cells on Insights stay empty until a real receipt is pasted in.

    What is the difference between Generation tok/s and full-turn tok/s?

    Generation tok/s is the decode-ish clock when Slowtest has a real generation window: own-key first-to-last token, or 1-paste start/finish beacons. Full-turn tok/s is the whole job at the keyboard — prompt prefill, hidden reasoning, tools, queue, UI slack, and the telemetry POST. A faster full-turn is not a smarter model. Think time is not Generation.

    Can I test AI speed without an API key?

    Yes. Send your agent to the test page, copy test instructions into your AI session, or use the manual prompt and timer. The optional direct API path uses your own key; local Ollama needs no cloud key.

    What does Slowtest fidelity measure?

    Fidelity is format and negative constraints, not quality. Line 1 must declare MODEL:. The essay must land between 800 and 1200 words on the first try — about 1000 words, one pass, no rewrite. A slow run at 100% fidelity is still a pass. A fast run that skips Line 1 or overshoots 1200 words is a miss. Speed with a miss is not a score.