The Forever Free API

The Forever Free API

Banana Dev logo on gradient background.

Hacker types! I know you’re here to get your hands dirty with GPTJ and CLIP so jump straight to using the free APIs. For those who do want to hang around, here’s a story:

Why we’re (re)launching the Free API

In 2020, Banana was a one-man team. It was right in the wake of the GPT3 boom, and I was itching to get this amazing technology out to the masses. So I hacked together a free GPT2 API and put it out into the world.

At that time, the resources were scarce. I was bootstrapped, with a shoestring budget. And as a solo engineer, I desired to sleep through most nights without needing to rescue prod. The early month of the product was stressful but amazing. We saw our name and API popping up in random youtube tutorials, which was dope. We had inbound users begging for paid GPU version.

Through 2021, out of a need to chase revenue and prove ourselves as a now venture-funded startup, we dove deep into b2b deals, and eventually sidelined the free API.

But with that choice, some of the magic was lost.

Today, we’re proud to announce today a return to form, with the launch of our Forever Free APIs:

  • GPU backed.
  • Running custom CUDA kernels, faster than any models out there.
  • Served 99.99% uptime infra.
  • Prod-ready. Like, actually prod-ready.

During our last year of innovating on custom APIs, we’ve built a playbook and a team that can make this happen. We’ve made magic, and we’re bringing it to the masses.

Today we launch our two most popular model implementations, GPTJ and CLIP, with more public and in-house models (carrot foreshadow 👀🥕) on the way. We're starting on limited infra to manage costs, and hope to scale over our lifetime to drive queue times to zero.

If you’d like to join us and ship ML to prod today, get set up on the Forever Free API.

Start here.


Erik 🍌