The short answer
Whizi carries Kimi from Moonshot inside the same subscription as the rest of the 280+ model catalogue. Seven Kimi identifiers sit on the 10 credit rung, so a Kimi message costs 10 credits on every row below. Two of them are reachable on Pro, the rest need Powerhouse, and Starter reaches no Kimi row.
| Model identifier | Credits per message | Plan required |
|---|---|---|
moonshotai kimi-k3 | 10 | Pro |
moonshotai kimi-k2.5 | 10 | Pro |
moonshotai kimi-k2-thinking | 10 | Powerhouse |
moonshotai kimi-k2.7-code | 10 | Powerhouse |
moonshotai kimi-k2 | 10 | Powerhouse |
moonshotai kimi-k2-0905 | 10 | Powerhouse |
moonshotai kimi-k2.6 | 10 | Powerhouse |
No catalogue model reaches production without a rung: a build test fails if one does. And a free account is not model-gated, so it can open any of these rows inside its 7 message lifetime allowance.
Which plan opens which Kimi row
Pick the Kimi reasoning row on a Pro subscription and what comes back is HTTP 403, code tier_upgrade_required, message "Upgrade your plan to use this model." Five of the seven rows behave that way on Pro, for two different reasons.
Two are held back on purpose. The reasoning row sits in the frontier reasoning tier and the coding row sits with the coding specialists, the two categories reserved so the top tier has something to sell. Both categories and their membership are listed on the models page.
The other three are Powerhouse by omission rather than by a rule. Access is resolved by identifier: the gate checks the Starter set, then the agent set, then the Pro list, and anything it does not find returns "powerhouse". Nothing promotes a row automatically, so the next Kimi identifier the catalogue gains lands on Powerhouse until someone adds it to the Pro list by hand. Since all seven rows charge the same 10 credits, the gate and the price are unrelated questions on this family. Of the two rows Pro does reach, K3 is the Moonshot flagship, and it is named on the Pro paywall card beside Claude Sonnet.
Tiers stack by rank, so Powerhouse opens everything Pro opens. If a Kimi row shows as locked, when a model is unavailable covers what to do next.
| Plan | Price per month | Kimi rows reachable |
|---|---|---|
| Free | No paid allowance | Any text model, inside a 7 message lifetime cap |
| Starter | $15.99 | None |
| Pro | $29.99 | moonshotai kimi-k3, moonshotai kimi-k2.5 |
| Powerhouse | $49.99 | All seven above |
Context window and cost per answer
The Whizi Cost Index prices the following Kimi rows. A standard answer there is 1,000 input tokens plus 500 output tokens, priced from list rates read on 20 August 2026, so the figures are comparable across providers rather than being a Whizi charge.
| Cost Index row | Context window | Cost per standard answer | Per thousand answers | Credits |
|---|---|---|---|---|
| Kimi K2 0711 | 131K | $0.00172 | $1.72 | 10 |
| Kimi K2 Thinking | 262K | $0.00185 | $1.85 | 10 |
| Kimi K2.7 Code | 262K | $0.00246 | $2.46 | 10 |
| Kimi K3 | 1M | $0.0105 | $10.50 | 10 |
Kimi K2 Thinking is the median priced row of the whole index, which prices 100 OpenRouter rows across 29 providers. The index is a price comparison table and its row count is not the size of the Whizi catalogue.
A window is not a per-turn budget. The flat budget is 40,000 input tokens and 20,000 output tokens, and it applies to any catalogue model whose window is at least 93,000 tokens. Every priced Kimi row clears that line, which makes this family a clean demonstration of how little the window column decides. The four windows run from 131K to 1M, and all four rows still send the same 40,000 in and 20,000 out and still cost the same 10 credits. The 1M window on the K3 row buys room for a conversation to keep growing, not a million token prompt. Long threads are covered in when the context is too long.
Why every Kimi row costs the same 10 credits
One credit is one message on the house model. A rung is derived from a reference turn of 3,000 input tokens and 800 output tokens divided by that same turn on the 1x model, rounded up when a model falls between two rungs, because the cheap direction is the one that costs money. The legal rungs are 1, 2, 3, 4, 5, 6, 8, 10, 12, 15, 20, 25, 30, 40, 50, 60, 80 and 100, and a model may not be priced off the ladder.
Kimi does not follow that arithmetic. The family is pinned at 10 by owner policy, against cost-true rungs of 2, 3, 4 and 12 depending on the row. So on the cheaper Kimi rows you are charged above what the row costs to run, and on the dearest rows you are charged below it. The Claude Sonnet class is pinned the same way, at 10 against a cost-true 12, which is why a Kimi turn and a Claude Sonnet turn draw the same amount from the same allowance.
An identifier with no rung at all falls back to 100 credits, the top of the ladder, chosen so an unpriced model surfaces as a support ticket rather than quietly eating margin. Every Kimi identifier in the table above carries its own rung, so none of them can land there.
The full ladder and the models on every rung are in the credits reference.
What an allowance buys in Kimi turns
Credits are one shared allowance per billing period, and every Kimi row spends 10 of them per turn, which reduces the arithmetic on this family to a division by ten. Weekly billing is offered on mobile only, and not on every tier, so the weekly column below is the allowance a weekly period carries where that product is sold, not a plan every tier can buy.
| Plan | Credits per monthly period | Credits per weekly period | Kimi messages if spent only on Kimi |
|---|---|---|---|
| Starter | 400 | 100 | Not applicable, no Kimi access |
| Pro | 2,000 | 500 | 200 monthly, 50 weekly |
| Powerhouse | 8,000 | 2,000 | 800 monthly, 200 weekly |
Credits do not roll over: the balance sums only the rows carrying the current period key, so an unspent allowance is unreachable once the period turns over.
Credit pricing is also gated per platform, and the allowlist value in production today is the web client. On a 1 credit row the two systems come to the same thing. A Kimi turn does not: off the allowlist it is charged one message against the message allowance rather than 10 credits against the credit allowance, so a Pro account is metered at 800 messages instead of 2,000 credits.
Switching into and out of Kimi mid-conversation
The charge follows the model that answered: creditsForModel returns a fixed integer per model identifier. Sending three messages to a 1 credit model and then one to Kimi in the same thread spends 3 plus 10, not 40. Switch back and the next turn is at the cheaper rate again.
What carries across a switch is covered in switching models mid-conversation.
What a Kimi row will not do
Auto will not route to it. Auto has six fixed targets and Moonshot supplies none of them. What it does have is a code target priced at 10 credits, the same rung every Kimi row sits on, so sending a turn through Auto is not a cheaper road to Kimi pricing than picking Kimi. Kimi stays a manual pick.
It will not generate media. Image, video and audio generation are separate picker sections with their own hand-curated rows.
It gets no special search behaviour. Web search is a per-request toggle, so a Kimi turn behaves like any other text model. See web search.
It carries no published capability flags. Whizi records no per-model vision, audio input or knowledge cutoff, so this page publishes none for Kimi.
- Seven Kimi identifiers sit on the 10 credit rung, and each costs 10 credits per message
- Starter reaches no Kimi row
- Pro reaches the K3 flagship and one K2 variant
- The reasoning and coding rows require Powerhouse
- A free account can open any text model inside a 7 message lifetime cap
- The four Kimi rows in the Cost Index all clear the 93,000 token threshold, so they get the flat 40,000 in and 20,000 out per-turn budget
- Auto never picks a Kimi row for you
Frequently asked questions
Which Kimi models are included in Whizi?
Seven Kimi identifiers carry a 10 credit price: the base K2 row, a dated K2 build, the reasoning variant, two further K2 variants, the coding variant, and K3.
What happens to a Kimi message when I have fewer than 10 credits left?
It is refused whole rather than part charged, and the app returns HTTP 429 with the code message_limit_reached and the message "Your monthly message limit has been reached." Because every Kimi row is 10, the refusal threshold is one number for the whole family: an allowance either has a Kimi turn left in it or it does not, and there is no per-row arithmetic to do first. A cheaper row whose rung fits the remaining balance still goes through. What to do when credits run out covers the rest.
Is Kimi cheaper than Claude in Whizi?
Not to you: both draw 10 credits a message, so a Kimi turn and a Claude Sonnet turn cost a subscriber exactly the same. They are not the same to run. In the Cost Index the Kimi reasoning row answers a standard question for $0.00185 against $0.007 for Claude Sonnet 5, and the top Kimi row at $0.0105 matches the older Sonnet rows to the tenth of a cent.
Where is Kimi in the model picker?
Under the Moonshot vendor logo rather than the generic fallback icon. Moonshot is one of the families the picker gives its own logo to. It is not in the Recommended section: that section is Whizi AI, Auto, one GPT row, one Claude row and one Gemini row, and it is constrained to models reachable on Pro or below. The other five sections are Agents, Image Generation, Video Generation, Audio Generation and All models.
Can I try Kimi without paying?
Yes, seven times. A free account is not model-gated and its credit allowance is 0, so a Kimi turn on the free tier spends a message off the lifetime cap and no credits at all. The same turn takes 10 credits off a Pro subscriber, so spending the whole free allowance on this family is 70 credits of Pro usage for nothing. The cap never resets, and once it is gone the account gets HTTP 402 with the message "Your free messages are used up. Start a subscription to keep chatting."