Less expensive options than Grok

[aholmes-think]

Any chance any other stereo mode supported options are coming our way that are less expensive than grok? We haven't had the need much, but we recently signed a client which requires the stereo mode for ai receptionist transcriptions. They have fairly heavy call volume, but not enough to warrant a dedicated server however, their grok transcription bill is hefty at roughly $40 to $50/mo! I've been trying to search for answers so I didn't ask a dumb question here, but alas, my search foo has not been great to find the answer I'm looking for.

If there isn't a way to do this already, then I guess I would turn this into a feature request. Maybe we would split the model utilization between transcription with one model and sentiment etc with another? Or perhaps since we have stereo mode, the channels could be split to independent files and then the transcription weaved back together based on time stamps? Just some late night ideas!

 

[KyriacosS_3CX]

Hello,

When you say heavy call volume, how heavy? Because with U10 the cost of a dedicated AI server drops a lot, and you could even share a more powerful on with different customers, so the full cost does not go to just one customer that is only using a percentage of its capabilities.

https://www.3cx.com/blog/docs/cloud-vs-local-ai-server/

 

[aholmes-think]

Hello,

When you say heavy call volume, how heavy? Because with U10 the cost of a dedicated AI server drops a lot, and you could even share a more powerful on with different customers, so the full cost does not go to just one customer that is only using a percentage of its capabilities.

https://www.3cx.com/blog/docs/cloud-vs-local-ai-server/

If my math is right roughly 25,000 minutes was $70. At the moment its just the one customer and why the math doesn't work well to do a dedicated server. Everyone else is on openai whisper and for my heaviest volume client its less than $4/mo.

 

[accentlogic]

@aholmes-think the jump from whisper transcription to full AI transcription with Summary and Sentiment is huge. We're giving out Grok at not charge on some smaller systems while we sort out billing for it - and they are consistently $75 to $125 per month.

I think this will be a pay now or pay later deal. If they are using the data it has a lot of value. $40 or $50 should not be a big deal until your own server is justified.

 

[aholmes-think]

@aholmes-think the jump from whisper transcription to full AI transcription with Summary and Sentiment is huge. We're giving out Grok at not charge on some smaller systems while we sort out billing for it - and they are consistently $75 to $125 per month.

I think this will be a pay now or pay later deal. If they are using the data it has a lot of value. $40 or $50 should not be a big deal until your own server is justified.

Agreed - in this deal we didn't know the costs so I rolled the dice and lost. Luckily $50/mo isn't the end of the world to get burnt on. I'm not sure the reasons or tech limiations, but I'm sure my $60/mo fireflies account which does a few dozen hours is using something at a much lower cost that grok.

 

[JamieR]

I'm interested in this given I've just moved a client who currently burns through around $250 USD a month on OpenAI Whisper over to Grok as they wanted speaker diarisation so we will see how the numbers stack up.

I did try and force 3CX to use an OpenAI model that included speaker diarisation but failed as I couldn't pass it a chunking policy so moved to Grok instead. I'd personally like to see 3CX adopt this natively to give an additional choice.

 

[KyriacosS_3CX]

Hi, a 250 USD usage is definitely in the ballpark to consider a standalone dedicated 3CX AI Server, you can easily trial it on a rented GPU for a few days to see how it fits with the clients needs and expectations.