Menu

Post image 1
Post image 2
1 / 2
0

Three of my agent's API calls were Opus. My logs said "200 OK" eight times.

DEV Community·Ansh Saxena·25 days ago
#VNK9fiwb
#agents#api#llm#monitoring#fullscreen#agent
Reading 0:00
15s threshold

If you run a multi-agent workflow — LangChain with fallbacks, CrewAI with different models per agent, AutoGen, or anything where someone (maybe past-you) configured model routing — this post is for you. Here's what the logs showed: [agent] Starting document analysis... [llm] Response received (200 OK) [llm] Response received (200 OK) [llm] Response received (200 OK) [llm] Response received (200 OK) [llm] Response received (200 OK) [llm] Response received (200 OK) [llm] Response received (200 OK) [llm] Response received (200 OK) [agent] Task complete. Enter fullscreen mode Exit fullscreen mode Eight successes. Nothing to investigate. Here's what actually happened: Model Calls Cost (USD) ───────────────────────────────────── claude-opus-4-6 3 $3.2325 claude-sonnet-4-6 3 $0.2775 claude-haiku-4-5 2 $0.0092 ───────────────────────────────────── Total $3.5192 Enter fullscreen mode Exit fullscreen mode Three calls to Opus. 92% of the bill. The model= config said Haiku.…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.

Read More