Compiler chooses Claude 4.1 Opus instead of much better and cheaper Claude 4.5 Opus

[mpopst]

I am running the VM using Claude hosted at AWS Bedrock. I wrote

agent planner:
model: opus
prompt: "You are the sprint planner. Follow instructions provided in your context."

It chose Claude 4.1 Opus, which would have been a disaster, given I am paying for token. The solution Claude suggested me is:

config models:
opus: claude-opus-4-5-20250514 # Opus 4.5 for complex reasoning
sonnet: claude-sonnet-4-5-20250514 # Sonnet 4.5 for coordination/validation
haiku: claude-haiku-4-5-20250514 # Haiku 4.5 for simple logging tasks

[mpopst]

Apparently the suggested syntax was an hallucination.

[mpopst]

How to reproduce.

Program:
session "Report your exact model ID"
model: opus

Model response:

● OpenProse VM Initializing...

📋 Parsing: Ralph/test.prose

session "Report your exact model ID"
model: opus

Program Structure:

• 1 session block with prompt: "Report your exact model ID"
• Model specification: opus

---

🚀 Executing Session: "Report your exact model ID"

● Task(Report model ID) Opus 4.1
⎿  Done (0 tool uses · 19.5k tokens · 10s)

● ---
✅ Session Complete

Result:
Model ID: us.anthropic.claude-opus-4-1-20250805-v1:0

---

🏁 Program Execution Finished

Session	Status	Model Used	Output
"Report your exact model ID"	✅ Complete	opus	us.anthropic.claude-opus-4-1-20250805-v1:0