ALL SYSTEMS LIVE·LLM LATENCY · MEASURED LIVE · NEUTRAL WITNESS
DATA WITH RECEIPTS · ED25519 · RFC 3161·--:--:-- UTC
Live · measured by us · signed

Which AI provider is fastest right now?

Provider speed changes by the hour, so every comparison chart is stale the moment it is posted, and you cannot verify the numbers anyway. This one is different. It is live, we measure it ourselves, and you can sign a snapshot and check it in your own browser.

API latency leaderboard

loading live measurements…

ProviderAvg latency (24h)UptimeSamples
measuring…

Cheapest models, by input price

loading live pricing…

ModelProviderInput / 1MOutput / 1MContext
loading…

China's open models, trending now

ranked by adoption on Hugging Face (likes and downloads), not speed. DeepSeek also sits in the latency board above.

ModelLabLikesDownloads
loading…

Why this one is worth trusting

the two problems with every other comparison, solved

Live, not a screenshot

We re-measure provider latency continuously. The board above is minutes old, not last month's chart. Speed changes hourly, so the feed does too.

✍️

Yours to verify

Sign a snapshot below and the result is an Ed25519-signed record. Check the signature in your own browser, against our published key. You do not have to trust us.

🗄️

On the record

These are our own measurements with sample counts and uptime, kept over time. So you can show what was fastest, and cheapest, at any past moment, not just now.

Sign this snapshot, then verify it yourself

the answer to "nobody knows if this is reliable"

Take the live board above, sign it, and you get a timestamped record anyone can re-check. Flip a single byte and the signature fails. That is the difference between a number on a slide and evidence.

Honest about the numbers. Latency here is our own probe from our infrastructure, so it is a relative live signal, not a guarantee of what you will see from your region, your model, or under your load. We sign what we measured and when. We do not claim it is the only truth, and provenance is not the same as being correct.

Call the feed yourself

keyless over MCP, or one POST over REST

curl -s https://dynamicfeed.ai/v1/batch \ -H 'content-type: application/json' \ -d '{"calls":[{"tool":"ai_provider_latency","args":{"window_hours":24}}]}'

Same data your agent can pull at decision time, with the provenance envelope on every datapoint. Need a key for higher limits? Grab one here. Want to embed the board? there is an iframe widget.

Dynamic Feed

Live data your agent can prove it acted on.

Speed, pricing, weather, hazards, outages, and more. The current public catalogue uses a consistent provenance envelope, with responses signable and re-verifiable. This leaderboard is one of them.

See it for agents →