Kimi Pricing and API Access for Builders
编辑标准与来源政策: 编辑标准, 团队. 内容均链至原始来源，见 方法论.
Kimi pricing is useful only when it is tied to a real workflow. Start with the official Kimi API docs, model list, pricing pages, and account limits, then compare the cost against the exact task you want to move. For builders, the decision is not whether Kimi is broadly cheap or powerful. The decision is whether Kimi improves one measurable workflow without creating unclear access, billing, or review risk.
Quick Decision
| Question | Check first | Builder decision |
|---|---|---|
| Is Kimi pricing API-based or subscription-based? | Kimi API pricing and recharge pages | Treat API use as token-cost planning; use ChatGPT business plans only for workspace subscription comparison. |
| Which Kimi model should be tested first? | Kimi model list and pricing table | Use the coding model for coding-agent tests; use lower-cost general models for summaries and routing. |
| Is there a usable free tier? | Official platform account, recharge, limits, and promotion pages | Do not promise free production capacity unless the official account page confirms it for the current workspace. |
| When should a team avoid switching? | Docs gaps, missing budget owner, unclear limits | Stay on the current provider until the cost and failure mode are measurable. |
Current Official Snapshot
Last checked: 2026-07-09. Use Kimi model list and Kimi pricing as the current source of truth before quoting any model name or budget. The official Kimi pricing navigation currently separates Kimi K2.7 Code, Kimi K2.6, Kimi K2.5, Moonshot V1, BatchJob, WebSearch, recharge, and rate-limiting pages. The pricing page also announces Kimi K2.7 Code as a coding-focused model family and points users to the model-specific pricing tabs.
Do not treat this page as a fixed price table. Treat it as a builder workflow for reading the official price table: record the model, whether the task uses input tokens, cached input, output tokens, web search, batch jobs, or recharge limits, then estimate the cost per accepted result. A headline price is not enough if the task produces long outputs or needs repeated retries.
| Official surface | What to capture | Why it matters |
|---|---|---|
| Kimi model list | Exact model name such as Kimi K2.7 Code, Kimi K2.6, or Kimi K2.5 | Prevents comparing an old blog post against the current model menu. |
| Kimi pricing page | Input token, output token, cache, batch, and web search units where shown | The useful budget is the cost of one accepted workflow, not one demo call. |
| Recharge and rate limiting | Account balance, quota, rate limit, and workspace approval path | A model can be attractive while the account path is still blocked. |
| RadarAI Kimi tracker | Why this model is on watch or test | Keeps pricing connected to a builder decision instead of a generic model ranking. |
How to Estimate a Kimi Workflow
Use Kimi API docs, Kimi model list, and Kimi pricing together. For each candidate workflow, write one row with the document length, expected answer length, retry rate, review time, and fallback model. Then run the same task on your current baseline. Kimi is interesting only if the accepted result is cheaper, faster, easier to review, or better suited to Chinese-English material.
For an archive-scale document review, imagine 120 bilingual vendor PDFs, each 8,000 to 20,000 tokens after extraction. The team wants a routing memo: vendor name, product area, API claim, pricing clue, and whether a human should inspect the original. Kimi may be worth testing because long-context reading and Chinese-English routing are exactly where token shape matters. The cost model should include extraction failures, duplicate paragraphs, long output summaries, and the reviewer's time to verify five sampled documents.
For a codebase explanation task, measure repository size differently. A cheap model call can become expensive if it repeatedly reads irrelevant files or produces broad explanations that do not help a reviewer. The better Kimi test is one module, one question, one expected answer shape, and one rollback decision: keep the explanation, ask for a patch, or stop.
Archive-scale Document Review Example
A founder has 600 pages of Chinese and English supplier material and wants to know which documents mention API access, on-prem deployment, model pricing, or compliance restrictions. The first pass should not ask Kimi to write a polished report. It should ask for a structured routing table with citations back to filenames and sections. The acceptance test is simple: sample 20 rows, open the original documents, and check whether the extracted claim is actually present.
That example exposes the real pricing question. If the output is short and reviewable, a higher input-token workload may still be acceptable. If the model writes long prose that a human must clean up, the budget gets worse even if the official token price looks attractive.
Cost Planning Checklist
- Record the exact model name from the official Kimi model list.
- Record the pricing unit from the official pricing page.
- Estimate input tokens, cached input if applicable, output tokens, web search, batch jobs, and retries.
- Add human review time because a cheap call is not cheap if it creates repair work.
- Decide whether the workflow is watch, test, adopt, or skip.
Failure Modes
The common failure is comparing model names without workload shape. A second failure is treating a launch promotion as durable budget. A third failure is ignoring output length: an apparently cheap input model can become expensive if it produces long answers that still need repair. Use Kimi when the official docs, account path, and measured workflow all line up. Keep it on watch when only one of those three is true.
Source Checklist
FAQ
Is Kimi free for builders?
Treat free access as account-specific until the official Kimi account or pricing page confirms what applies to your workspace.
Should Kimi replace an existing API provider?
Only after one measured workflow beats the current baseline on cost, quality, and review time.
What should I compare first?
Compare one coding or long-context task with fixed inputs, expected outputs, and a retry budget.
Can I cite third-party price summaries?
Use them for discovery only; cite official Kimi pricing and model pages for decisions.