kimi-for-coding: inherit indicative costs from base models instead of zeroing

[lorecanc]

Summary

The kimi-for-coding provider models currently ship with zeroed [cost] tables, with the comment "Cost is zeroed per this provider's subscription convention". This makes usage tracking in opencode (and any models.dev consumer) report $0.00 for all Kimi For Coding traffic, even though the models are billed on the underlying model's token economics.

This PR removes the zeroed [cost] tables so each model inherits the cost of its base_model, keeping the pricing in sync automatically when the underlying model or its price changes.

What changed

• providers/kimi-for-coding/models/kimi-for-coding.toml — no longer zeroes cost; inherits from moonshotai/kimi-k2.7-code (input $0.95 / output $4.00 / cache read $0.19)
• providers/kimi-for-coding/models/kimi-for-coding-highspeed.toml — inherits from moonshotai/kimi-k2.7-code-highspeed (input $1.90 / output $8.00 / cache read $0.38)
• providers/kimi-for-coding/models/k3.toml — inherits from moonshotai/kimi-k3 (input $3.00 / output $15.00 / cache read $0.30)
• providers/kimi-for-coding/models/k3-256k.toml — inherits from moonshotai/kimi-k3

Each file now carries a comment clarifying that Kimi For Coding is a fixed-fee subscription plan and the inherited cost is the indicative pay-per-token equivalent, for usage tracking only, not the actual billing rate.

Rationale

The kimi-for-coding model IDs are rolling aliases that point at the current flagship coding model (e.g. kimi-for-coding → moonshotai/kimi-k2.7-code today). Because the models use base_model, dropping the zeroed [cost] tables yields:

• Realistic cost estimates in usage dashboards instead of $0.00
• Automatic price updates when the underlying model or alias changes
• No maintenance of hardcoded numbers per alias

This follows the existing convention for other subscription-billed endpoints already listed in this repo — e.g. providers/opencode-go/models/kimi-k2.7-code.toml ships full per-token costs (input $0.95 / output $4.00 / cache read $0.19) without zeroing.

Notes

• No schema changes; this is purely data.
• If maintainers prefer keeping subscription endpoints at $0, an alternative would be a dedicated indicative flag in the schema — happy to discuss.

[github-actions]

Action items

• [high] [violation] providers/kimi-for-coding/models/kimi-for-coding.toml:1 - Check: Provider models must author cost themselves; base_model only merges lab metadata under models/, and cost is provider-only. Why: Removing [cost] does not inherit Moonshot pay-per-token prices. Lab files models/moonshotai/kimi-k2.7-code.toml, kimi-k2.7-code-highspeed.toml, and kimi-k3.toml have no cost, so these four models resolve with no cost (not $0.95/$4.00 etc.). The PR body is incorrect on that point, and the cited providers/opencode-go/models/kimi-k2.7-code.toml authors [cost] explicitly rather than inheriting it. Action: Restore provider-authored [cost] on all four files. Either keep zeros (subscription convention used by other coding-plan providers) or copy the intended indicative USD/MTok numbers from providers/moonshotai/ into each file; do not rely on base_model for pricing.