Profiled model server performance, identified hardware limit — 737.5k tokens

Profiled model server performance, identified hardware limit

agent-harness

LOCKED IN

First session on agent-harness

First build tracked for this project

737.5ktokens

9prompts

11:24:15time

1,667lines

Claude Codeclaude-opus-5Markdown$29.07

mostly building

Investigated configuration options exhaustively and confirmed the server's performance is hardware-limited.

Tokens / prompt81.9k

Cost / line$0.017

Cache hit99%

Burn rate$2.55/hr

Derived from this session's token and cost data. Not shown on the feed.

Building 82%Verifying 18%