Models are becoming smarter, faster, and cheaper, but the OCR industry hasn’t kept up.
We’re launching r-1 mini: the first model to offer truly premium parsing without the premium pricing. At $1/1000 pages, r-1 mini beats options that cost up to 25x as much.
A really big part of starting a company is having a deep-rooted belief that you can create your own circumstances.
I credit that mindset to my parents, who I watched repeatedly overcome challenges after immigrating to the US.
Getting to work on @reductoai with the incredible people on our team has been fulfilling in ways I couldn’t have guessed, and @martin_plural nailed what it feels like behind the scenes. Think this is my most personal interview.
https://t.co/Y2ly2OMcLh
What's it like to be a young founder in the age of AI?
We asked Applied Compute's @ypatil125, Gumloop's @MaxBrodeurUrbas and Reducto's @aditabrm: ever-expanding competition, anxiety while not working, replacing inexperience with intuition.
https://t.co/mVFrDV9A20
the way sf literally broke a 108-year-old rain record and we were still running around hand-delivering reducto goodies to our favorite startup partners ☔️💜
This test shows why you can't just trust benchmark leaderboards, and it's the simplest test we ran.
- we ran the native bench
- we ran with PAIR
- we ran with 2x super simple ablated prompts
- we ran with fable, astra, argon
The reality is more extreme. Build your own evals!
Last week, I said every parser is #1 on its own benchmark. So we tested the benchmarks.
We ran 8 parsers across 12 benchmarks, then tilted the pages by 2–4°. 6 benchmarks got a different winner.
We’re calling it BenchBench.
respectfully Jerry, we both know that the aggregate scores are presented as a measure of parsing accuracy that's supposed to align with human judgement
I'm not even upset about the example on our behalf. Reducto's result (on the right) is graded as correct, and the table rated as 0 is from another competitor of ours. I personally think their output here is fine from the lens of what customers should care about.
I don't think it's right to say that this isn't a problem because it's a formatting test, but if it helps here's an example of a table test: https://t.co/J7bhruTKil