Kimi K3 recurring-plan comparison
Which plan tells you what you’re actually buying?
The cheapest price is easy to publish. The quota is not. This lab separates exact K3 allowance, relative limits, and “unlimited” claims with restrictions—without mixing in PAYG-only APIs.
- Scope
- Recurring plans with explicit K3 access
- Removed
- PAYG-only providers and K2.x-only offers
- Strongest
- OpenCode Go publishes K3 request estimates
- Weakest
- Five major plans hide absolute K3 capacity
Price is visible. Quota honesty is the differentiator.
Higher means the vendor publishes more of the K3 allowance: exact request windows or convertible credits at the top; relative multipliers and unnamed pools near the bottom. Circle size indicates exposed context.
How far can the published allowance go?
Blue is the maximum if every included dollar or credit becomes uncached input. Orange is output-only. Actual coding sessions mix input, cached context, reasoning, and output, so neither bar predicts a blended workload.
Maximum K3 token equivalents
MILLION TOKENS · ENTRY TIER“Unlimited” survives only after the footnotes.
One plan explicitly offers unlimited K3-capable tokens. It is not an unlimited autonomous coding plan.
Featherless is unlimited*
Choose the evidence you need, not the logo you know.
Each recommendation follows a hard requirement. Where the quota is opaque, the board says so rather than manufacturing a token estimate.
Every confirmed recurring plan.
Use the filters above the market map; the same filter dims rows here. “Transparent” means the K3 allowance can be inspected or converted—not that the plan is necessarily a better value.
| Service | Entry price | K3 allowance | Context evidence | External use | Honesty | Primary caveat | Source |
|---|
Primary sources, not provider roundups.
All linked pages returned HTTP 200 during verification. Authenticated CLI catalogs were queried without sending model requests or consuming inference quota.