Support All OpenRouter Models

[JoshuaKirby88]

Summary

Adds OpenRouter model support across the AI Scientist pipeline, including standardized model naming/normalization and a shared CLI validation helper so users can pass openrouter/<provider>/<model> without editing AVAILABLE_LLMS.

What Changed

• OpenRouter model routing + normalization

  • Adds support for openrouter// model names across the pipeline.
  • Keeps backward compatibility with the previously supported llama3.1-405b alias (no openrouter/ prefix) by normalizing it to meta-llama/llama-3.1-405b-instruct.
  • Uses the same OpenRouter OpenAI-compatible client config as before, now applied to all OpenRouter models instead of a single hard-coded case.

• Reusable CLI model validation

  • Introduces validate_model_choice() (argparse type=) to accept either:
    • models in AVAILABLE_LLMS, or
    • anything starting with openrouter/<provider>/<model>
  • Updates CLI argument parsing in the main entrypoints to use this validator instead of choices=AVAILABLE_LLMS:
    • ai_scientist/generate_ideas.py
    • ai_scientist/perform_writeup.py
    • launch_scientist.py

• Writeup / Aider model selection

  • Replaces the one-off llama3.1-405b handling with generalized OpenRouter handling when constructing aider.models.Model(...) in:
    • ai_scientist/perform_writeup.py
    • launch_scientist.py

Usage

• Existing models in AVAILABLE_LLMS still work unchanged.
• New supported format:
  • --model openrouter/<provider>/<model>
  • Example: --model openrouter/meta-llama/llama-3.1-405b-instruct