Hi everyone, I was testing Claude in ucode , and all my sessions were starting with 30% of context window already used.
I can see that the Opus 4.7 used in Databricks vibe agent (left) has a 1M context, and the AI Gateway Opus 4.7 (right) has a 200k context.
But still the initial context was ~37k tokens in the first, and ~59k in the second.
I sent this to #ai-gateway slack channel and was prompted to create an issue about it.

Hi everyone, I was testing Claude in
ucode, and all my sessions were starting with 30% of context window already used.I can see that the Opus 4.7 used in Databricks
vibe agent(left) has a 1M context, and the AI Gateway Opus 4.7 (right) has a 200k context.But still the initial context was ~37k tokens in the first, and ~59k in the second.
I sent this to
#ai-gatewayslack channel and was prompted to create an issue about it.