extended-salmon•16mo ago
@everyone we've dropped the cost of AI
@everyone we've dropped the cost of AI tokens by 50%!
In the background we're actually working on making AI tokens at-cost (as a part of a big billing update).
It's taking longer than we initially hoped, so in the meanwhile we've slashed the price of tokens in half while we work on this!
14 Replies
national-gold•16mo ago
Let me know if you have any question 🙂
other-emerald•16mo ago
legends
Appreciate this VF team!
quickest-silver•16mo ago
@Sohail C
fair-rose•16mo ago
This is awesome 🙌
quickest-silver•16mo ago
Great news, thanks 🔥
genetic-orange•16mo ago
You just made me go from MVP to paying rent thanks
national-gold•16mo ago
HAHAHA love that!!
adverse-sapphire•16mo ago
Amazing 👏
conservation-jade•16mo ago
bravo! true story: I called out another service we use last week (essentially a GPT wrapper) when they actually raised their AI prices. I said OpenAI halved their prices with turbo, and halved them again with gpt-4o, why are you increasing prices instead of reducing them? They literally laughed at me. will be advocating for using VF even more. thank you!
genetic-orange•15mo ago
Hey Daniel, follow up question. As devs building on a third party SaaS it's important for us to know what this long-term relationship looks like. From what I see, VF monetizes through 3 streams: AI tokens, seats for team work, and enterprise support.
The idea of bringing to AI tokens to cost is amazing for a small agency as mine, but I'm worried if that means that your business model will change in the future in unexpected ways.
This move makes if anybody scaling on VF was already bypassing your AI tokens and just doing custom API calls, or moving to a competitor, which honestly, makes sense. But I hope that the revenue from AI agents was a small percentage of overall revenue or just impossible to sustain long term that you decided to make the switch now.
Should I assume then that VF's revene growth as a platorm will primarily be team & enterprise-driven and smaller clients like me shouldn't expect sudden changes in the future?
national-gold•15mo ago
Hey @spaceman212 great question, we monetize in 2 ways
1. Software Subscription
2. Production Usage
AI tokens was never a significant revenue source, it was mainly just legacy to ensure all costs were being covered as we added and experimented with new AI features.
For #1 this is what everyone pays for their plan. Largest amounts are enterprise ofcourse.
For #2 we've been doing this manually thus far with customers that have 250,000+ interactions a month (roughly translated to 25,000+ conversations a month). This only is relevant for roughly 1% of the user base that is doing this volume. Usually at the teams or enterprise level. This is where we incur server costs.
We will be moving to an automated/self-serve system for #2, but this only comes into effect once you start doing significant production volume. This will only impact that top few % of customers.
As an agency owner its important that you have some mechanism for charging customers based on volume as thats where costs are incurred.
genetic-orange•13mo ago
Hey Daniel, just following up on this topic. What's the current ETA?
Right now an accumulated use of 1 million input tokens and 1 million output tokens on gpt4o mini would be $0.75 calling OpenAI directly, but at $5 per 2 millon tokens times 0.2, it's $1 on Voiceflow
Also, a bit of random question. Are people only building chatbots on Voiceflow? I've started working on a new project which is more about chaining LLM requests, saving variables (looong text strings) and doing decision trees based on IFs.
For example, let's say I scrape a website's text to create a summarized description of a company (long strings). But if it's ecommerce I want it to read one way and if it's a SaaS I want to read it another way.
Theoretically this could be done on Voiceflow, but I'm not sure if this is the right tool or if I'll come across issues if your backend is designed for short chat-like messages.
national-gold•13mo ago
No ETA - there's a number of other roadmap items we're trying to get too first that have become higher priority as this involves quite a bit of systems work.
And great question! You can definitely use Voiceflow in the way you're describing (aka. visual, more complex, Zapier style chaining). You can even kick those off using the DM API - Intent Request. People build a range of agents on Voiceflow and thats one type of them.