Model Normalization

basemode aggressively normalizes model names before strategy detection and API calls. This lets you use short names, aliases, and provider-prefixed forms interchangeably.

What normalization does

1. Resolves aliases — short names like claude-opus-4-7 expand to anthropic/claude-opus-4-7
2. Infers provider prefix — gpt-4o-mini becomes openai/gpt-4o-mini
3. Fixes Anthropic name format — dots become dashes (4.6 → 4-6)
4. Passes through already-normalized names unchanged

Examples

Input	Normalized
gpt-4o-mini	openai/gpt-4o-mini
claude-opus-4-7	anthropic/claude-opus-4-7
claude-3-5-sonnet	anthropic/claude-3-5-sonnet-20241022
gemini-2.0-flash	google/gemini-2.0-flash
deepseek-coder	deepseek/deepseek-coder
anthropic/claude-opus-4-7	anthropic/claude-opus-4-7 (unchanged)

Usage in code

from basemode.detect import normalize_model

model = normalize_model("claude-opus-4-7")
# → "anthropic/claude-opus-4-7"

Normalization happens automatically inside continue_text() and branch_text() — you rarely need to call it directly.

Provider inference

When a model name has no provider prefix, basemode scans a list of known model name fragments to determine the provider:

• gpt, o1, o3, davinci, turbo → openai
• claude → anthropic
• gemini, gemma → google
• llama → together (or groq depending on context)
• deepseek → deepseek
• mistral → mistral
• etc.

If detection fails, the name is passed through as-is and LiteLLM attempts its own resolution.