everyone has talked about a fennec, but nobody seems to talk about a capybara…

i wonder if anthropic will discontinue haiku and replace it with sonnet

this explains a lot

wen yu ar claude butt undercover

?..

looks like anthropic started internally testing capybara-v9-fast[1m]. i don't have direct access to [ANT ONLY] model overrides, but it appeared today

there is also a mention of "ClaudeWatch", not yet entirely sure what it is

I'M NOT SURE, but this appears to be some internal model monitoring and/or evaluation system?
from what I see it appears to be tracking some specific model behaviors and user answers to them (positive/negative)? it also seems to build metrics depending on that. potentially a very small scope of what it is or what it does

btw, fun fact, opus/sonnet 4.6 fallbacks to sonnet 4. just 4. the model released a year ago. i didn't even know they still run it

and i feel like this might be happening quite often, potentially even on API level?
would love if anyone could share their metrics from CC sessions to see if it uses sonnet-4 directly

potential new model codenames? comes from a bit unverified source but he seemed to know about -v9

bandicoot
bandicoot-v5

canele
canele-eap
canele-v1

Disclaimer: I am not affiliated with, endorsed by, or in any way officially connected to Anthropic PBC or any of its employees, officers, or agents. All information shared in this channel has been obtained exclusively from publicly available sources, including but not limited to public repositories, social media posts, publicly accessible documentation, and other open-source intelligence. No proprietary, confidential, or trade secret information has been knowingly disclosed. All analysis and commentary represents my own personal interpretation and speculation. Nothing shared here should be construed as insider knowledge, official communication, or endorsed information from Anthropic PBC. I reserve the right to remove any content at my sole discretion.

let me present you

linkedin llm arena