A dev's $200/mo limit drained in 2 days.
He pulled the logs: 145M tokens re-reading old context. 5.9M doing real work.
93% of the bill was the agent re-reading itself.
Only watching your token counter isn't cost control.
With Carbon, take complete control of your AI system.
An AI coding agent ran for one day for personal use. & Bill: $788
78% went to a single model — doing work a cheaper one does for $1.70. Same output. ~360× the price.
You're probably doing the same thing right now. You just can't see it.
Check yours→ https://t.co/ItdbtYpC4F
We’re launching Carbon in open beta.
Carbon is the analytics layer of AI.
Companies are shipping AI features faster than they can understand them. We turn your AI data into a queryable ledger — so engineering and product decisions stop running on guesswork.
Why we built it 👇
I am pretty sure OpenAi would have launched 5.6 by now.. and maybe when they were about to then this fable fiasco happened 🤦♂️
Or maybe they would use this as an excuse in their next investor meeting 😂
Grok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come.
Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release.
This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.