#hey! claude has trciky latency issues

1 messages · Page 1 of 1 (latest)

high cove
#

thanks for followup! unluckly I was also testing with GPT 4.1 and it happens as well

here are some conv_ids:

  • conv_01jvdj9nqpfs0a1ee3hjy423zk
  • conv_01jvdj1j6mftjbe20zf0pngcqq
  • conv_01jvde1nn4f21beqrnrapgxg7k
  • conv_01jvddze1cfxgtps74q5ea0eap
  • conv_01jvd44f47f5qat38yrjskhn5e

If you need more examples I'm happy to provide as there are much more 😄

this one is for example valid as sometimes it works correctly:

  • conv_01jvdbjcfaeafrv8m95j62z6fd
high cove
#

conv_01jvf32bt1fa9an1jqmdsscr7k

here is another one just from today. You can hear that I start saying 'haha' and recording ends, while on phone call I was saying something. Call gets ended by Twilio
I also checked any my convesation limit is 600 seconds

high cove
#

testing it right now and all the time twilio errors with: 31951Stream - Protocol - Invalid message

it happens usually around 40th second of call but sometimes sooner or later but eventually it fails :/
thanks for assistance @shell hamlet

#

fyi: I made new, pure agent and still happens when doing inbound calls with Twilio

high cove
#

probably this thread will be lost in the number of messages on the channel, so I gently bump it @shell hamlet 😛

left patio
#

Hello. I am having this same error. Were you able to get to the root of the problem?

high cove
#

thanks @left patio for joining thread, I am wondering how many people are experiencing it

shell hamlet
#

currently you're seeing 700ms time to first byte from LLM

#

that will cause higher latency

high cove
#

thanks @shell hamlet . Do you know what causes this ? Is this a Twilio integration with servers in US? 🤔

shell hamlet
#

i believe simply the fact that the claude agent is not yet specialised for realtime use cases as their model is designed to be the most intelligent possible

#

so their model is quite large

high cove
#

hmm, I was testing it with 4.1 model from open ai as well and gemini 2.5
I will get back on this next week, to test once again and let you know. Maybe it was a temporary issue.
Thank you for assistance!

onyx wave
#

@high cove Were you able to reduce the latency?