Use upstream per-token prices for subscription plan providers instead of $0

[zlliang]

Subscription plan providers currently represent reference model prices inconsistently. Most text models use an explicit $0, while OpenCode Go uses the underlying models' public USD pay-as-you-go prices even though it is also a subscription.

This makes paid plan models appear free in price comparisons and causes downstream cost tracking to report $0 for usage that consumes a paid quota.

Evidence

The same underlying model currently has different prices depending on the subscription provider:

Entry	base_model	input	output	cache_read
moonshotai/kimi-k2.7-code	—	$0.95	$4.00	$0.19
opencode-go/kimi-k2.7-code	moonshotai/kimi-k2.7-code	$0.95	$4.00	$0.19
kimi-for-coding/k2p7	moonshotai/kimi-k2.7-code	$0	$0	$0

OpenCode Go has reference prices for all 21 models. Across the 16 subscription plan providers below, only 5 of 107 token-priced text models are priced; 96 have an explicit $0, and 6 omit cost.

Provider	Explicit $0	Missing cost	Priced	Total
alibaba-coding-plan	10	0	2	12
alibaba-coding-plan-cn	10	0	2	12
alibaba-token-plan	14	0	0	14
alibaba-token-plan-cn	14	0	0	14
kimi-for-coding	6	0	0	6
kuae-cloud-coding-plan	1	0	0	1
minimax-coding-plan	7	0	0	7
minimax-cn-coding-plan	7	0	0	7
stepfun-ai-step-plan	0	3	0	3
stepfun-step-plan	0	3	0	3
umans-ai-coding-plan	6	0	0	6
xiaomi-token-plan-ams	3	0	0	3
xiaomi-token-plan-cn	3	0	0	3
xiaomi-token-plan-sgp	3	0	0	3
zai-coding-plan	6	0	0	6
zhipuai-coding-plan	6	0	1	7

There are also inconsistencies within a provider: alibaba-coding-plan prices two Qwen models while the other ten are $0, and zhipuai-coding-plan prices one GLM model while the other six are $0.

Requested change

Use the underlying model's public USD pay-as-you-go price for subscription plan text models, including documented context tiers and cache prices. This keeps the plan's billing model separate from the model's reference price.

A proposed implementation is available in #3304. I am willing to contribute and maintain the fix based on maintainer feedback.

[github-actions]

This issue is not an actionable model catalog data fix — it's a convention/policy proposal that explicitly asks for maintainer agreement before any changes are made.

The core of the request is:

• A policy decision on whether subscription/quota-plan providers should record upstream per-token list prices instead of $0.
• A README documentation change ("This should also be documented in README so future provider additions follow one convention.").
• A broad sweep across 19 providers and 139 models, gated on maintainer sign-off ("I am willing to contribute the fix once maintainers agree with this convention.").

While the issue cites specific price inconsistencies (e.g., kimi-for-coding/k2p7 at $0 vs opencode-go using upstream prices, or alibaba-coding-plan/zhipuai-coding-plan pricing only a subset of models), those are presented as evidence for the convention proposal rather than isolated factual errors with a single verifiable correction. Resolving them the way the issue proposes requires first deciding the convention (a maintainer policy call) and then sweeping the catalog — neither of which is a minimal, independently-verified data fix I can make safely without guessing.

Per the guidelines, I do not make documentation, workflow, or convention changes, and I do not edit files for policy/product discussions. No automated fix was opened.

Recommendation for the maintainer: Decide whether subscription-plan provider TOMLs should inherit the base model's public per-token price (by dropping zeroed [cost] overrides for base_model-bearing entries) or continue to encode plan access as $0. If the convention is adopted, the resulting changes would touch many provider TOMLs and the README, and should be coordinated as a single tracked effort rather than piecemeal.