v0.15.0

Added

• Azure GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna (azure-gpt-6.1-sol · azure-gpt-6-sol · azure-gpt-6-luna, Global standard): the
  OpenAI rates — $2 / $10 (cached $0.10 · $0.20), Luna $0.10 / $0.50 — with their long-context tier over 272K and cache writes.
• Mistral Large 4 (mistral-large-4, also mistral-large-4-0): $0.68 / $2.09, cached input $0.07, 1M context — Mistral's
  current sale price (list $1.36 / $4.18). Before, the id fell under the mistral-large alias and was priced as Mistral Large 3
  ($0.50 / $1.50).
• Codestral Embed (codestral-embed, embedding): $0.15 input. Before, codestral-embed-2505 fell under the codestral alias
  and was priced as the chat model ($0.30 / $0.90).
• Qwen3.7 Max (qwen3.7-max): $2.50 / $7.50, and Qwen3.6 Plus (qwen3.6-plus): $0.50 / $3.00 up to 256K input,
  $2.00 / $6.00 above (Model Studio, international).

Fixed

• A family alias no longer prices a newer version of the family. An alias ending in a family name (mistral-large,
  codestral) matched any id that went on with a version number — mistral-large-5 was priced as Mistral Large 3. One or
  two digits after a hyphen or underscore (ending the id or followed by a separator) now read as a version the catalog does
  not know, the rule aliases ending in a number already followed; a snapshot date (-2512) or a size (-17b) still matches.
  grok-code-fast-1 keeps resolving through an exact alias.

Changed

• command-r is marked deprecated (DeprecationDate 2025-09-15, ReplacementModelId command-r-08-2024) and named
  «Command R (03-2024)»: Cohere's alias resolves to the March 2024 model, whose legacy price ($0.50 / $1.50) the row keeps.