Revenue on our TinfoilAI-hosted TEE models is up ~20x since launch in March. TEE models make it so that nobody, not us, not the model host, can read your prompts.
Private inference won't be a niche. It's going to become the default.
Instead of spending time with your family this weekend, consider plugging the new DeepSeek V4.1 Flash 1017 Ultrafast into your AI agent!
Now available, it streams at a blazing 1500+ tokens per second.
deepseek-v4.1-flash-1017-ultrafast
PPQ's revenue is 15x what it was this time last year.
Thank you to our die-hard customers and their great feedback. It got us to where we are now and will continue to take us into the future.
Verifiable privacy, quality, and simplicity. That's the mission.
One of the largest Lighthing nodes (cyberdyne.sh) shut down. A real tragedy... Learned much from this operator over the last couple of years. @Silen Zara what happened with the node? ๐ญ๐ญ๐ญ
New guide: Strix + PayPerQ. A project with 70k Github starts that pentests your app. Swap in one of our new abliterated models when proprietary models refuse cyber questions.
Many customers asking how to red-team their own codebase.
We just added the
@abliteration_ai
models (GLM 5.3 and DeepSeek V4.1 Flash, abliterated)
Refusals are removed from the weights, so they work through your security issues instead of declining!
We've got internal access to this model and its FIREEEEEE.
We're told we'll be able to roll this out to users in a matter of a week or so. ๐ค๐ค๐ค
Does your harness ever encounter a javascript rendering or bot detection issue when using standard curl or webfetch attempts?
If so, you can tell your harness to use firecrawl
in our data enrichment endpoints to get the scraping done!
Example harness instructions:
**Blocked web pages โ escalation order:** when WebFetch or `curl` returns a bot-detection or "checking your browser" page instead of content, try in this order, stopping at the first that works:
1. Firecrawl through our own endpoint (`POST /v1/data/api/firecrawl/scrape`, body `{"url": โฆ, "formats": ["markdown"]}`, ~$0.015/call). It renders JavaScript and handles standard Cloudflare-style challenges, so it is the right second attempt for any JS-rendered or bot-gated page.
2. The site's own alternative front door: a `noscript.` subdomain or link in the challenge page, an onion/I2P mirror, an RSS/sitemap/`llms.txt`, or its public API.
3. A cached copy: `curl` the Wayback Machine (`https://web.archive.org/web/2026id_/<url>`, pass `--compressed`); WebFetch cannot reach
Do not loop on the same fetcher with longer waits โ one retry is enough evidence that it will not pass.
Head to our data enrichment page to learn more!
Google's image and video models are 95% off on PPQ for the next couple of weeks.
Nano Banana 2.1, Nano Banana Pro, Veo 3.1, Veo 3.1 Fast and Gemini Omni 1.1 Flash. A Veo clip for about 2 cents, an image for a fraction of one.
Pay-per-use, no subscription!
PS: Yes, it's real!
Gemini 4 Argon has arrived at PPQ. and we're slashing the price heavily:
75% off of the list price (50% off the price that even Openrouter charges)
$1 input and $5 output per 1M tokens
While supplies last! Go nuts!