Which models / price tiers should ship in the default MODELS table?

[cognis-digital]

Pricing in tokenmeter is data, not logic — the MODELS table in tokenmeter/core.py maps a name to (input_per_1k, output_per_1k, context_window), and you can register one at runtime with add_model(...). Today it ships Claude Opus/Sonnet/Haiku, GPT-4o / 4o-mini / 4-turbo, and a generic-1k.

What models and price tiers should ship in the default table? Drop a model name, its public list price (USD per 1K input / output), and context window and we'll add it. Tracking ideas here:

• Embedding models (priced per 1K, output-free)?
• Long-context tiers (1M-token windows)?
• Per-region / batch-API discount tiers as separate entries?

Keep tokenmeter models --format csv under version control and diff it after each vendor price change — this thread is where we coordinate those updates.