# Models a provider offers that evaling deliberately does not price.
#
# Anything here reports an unknown cost, which is the honest answer when we
# cannot source a rate. Anything NOT here and not in PRICES is a finding from
# tools/check_model_coverage.py — price it, or add it here with the reason.

# -- Not text models, or not callable through evaling ---------------------
# Base completion models: no chat interface, so evaling cannot call them.
babbage-002
davinci-002
gpt-3.5-turbo-instruct
gpt-3.5-turbo-instruct-0914

# Gemma is open-weights and usually self-hosted, so the operator sets the
# rate. Priced with params.pricing, like any local model.
gemma-4-26b-a4b-it
gemma-4-31b-it

# -- Moving aliases the card does not price -------------------------------
# These track whatever is current, and no card gives them a rate of their own.
# A rate guessed from today's target goes stale silently and under-counts once
# the alias moves to a costlier model. See the note in docs/providers.md.
#
# Where a card *does* price an unversioned name, it goes in PRICES instead —
# `chat-latest` is the one that does.
gemini-flash-latest
gemini-flash-lite-latest
gemini-pro-latest
gpt-4
gpt-4-turbo
gpt-5-chat-latest
gpt-5.1-chat-latest

# -- Offered but absent from the published rate card ----------------------
# The card is the only source for a rate, so a model missing from it cannot be
# priced without guessing. Re-check at each release: when one appears on the
# card, move it into PRICES.
gpt-5-codex
gpt-5.1-codex
gpt-5.1-codex-max
gpt-5.2-codex
gpt-3.5-turbo-16k
gpt-4o-search-preview
gpt-4o-mini-search-preview
gemini-3.1-flash-lite-preview
