Moonshot AI's Kimi API bills Kimi K3 per token from prepaid credit, pay-as-you-go, with no subscription. Moonshot now calls K3 its previous-generation flagship, behind the new Kimi K3.1 Pro, but it remains on sale. Here are the K3 rates, what common workloads cost, and the caching, batch and top-up rules that decide what you actually pay.
Kimi K3 costs $3.00 per 1M input tokens on a cache miss, $0.30 on a cache hit, and $15.00 per 1M output tokens, across a 1M context window with no context-length tiers. Batch jobs are billed at 60% of the standard price, and the full model weights are public if you'd rather self-host.
| Model | Tier | Input / 1M | Output / 1M | Cache hit / 1M | Context | Notes |
|---|---|---|---|---|---|---|
| Kimi K3.1 Pro | Flagship | $2.00 | $10.00 | $0.20 | 1M | Moonshot's current flagship, a further post-trained K3 at a lower price; cache write $2 (5-min TTL) or $4 (1-h TTL); Highspeed variant $4/$20 |
| Kimi K3 | Flagship | $3.00 | $15.00 | $0.30 | 1M | Previous-generation flagship, superseded by Kimi K3.1 Pro but still sold; open weights; cache write $3 (5-min TTL) or $6 (1-h TTL) |
| Model | Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens |
|---|---|---|---|
| Kimi K3.1 Pro | $80.00 | $68.00 | $3.02 |
| Kimi K3 | $120 | $102 | $4.53 |
Standard rates per 1M tokens from data on this site. The coding-agent row assumes 90% of input is read from the prompt cache (full input price where a model has no cache rate). Cache-write fees, batch discounts and long-context surcharges listed in the notes are not included. Updated November 2, 2026.
| Model | Provider | Tier | Input / 1M | Output / 1M | Context |
|---|---|---|---|---|---|
| Kimi K3.1 Pro | Moonshot AI | Flagship | $2.00 | $10.00 | 1M |
| Grok 4.7 | SpaceXAI | Flagship | $2.00 | $6.00 | 500K |
| Qwen3.8-Max | Alibaba | Flagship | $2.00 | $6.00 | 1M |
| Gemini 4 Argon | Flagship | $2.00 | $10.00 | 1M | |
| Gemini 3.1 Pro | Flagship | $2.00 | $12.00 | 1M |
Moonshot AI's cheapest model per tier (bold) against the four cheapest same-tier models from other providers, by output price.
Moonshot bills caching as three items. A cache write with the default 5-minute TTL is charged once at the normal input rate ($3.00 per 1M), and a 1-hour TTL write costs twice that. Every later hit is billed at $0.30 — one-tenth of a miss — and renews the entry. Moonshot's own maths says the 1-hour tier pays for itself after two extra hits within the hour, so pick it for agents or chat sessions with gaps longer than five minutes. Set it with prompt_cache_options.ttl; the TTL is fixed at first write. The coding-agent estimate on this page leaves cache-write fees out.
K3 is gated as a flagship model: it unlocks after a first top-up (the minimum is one US dollar), and your cumulative top-up total sets your account tier and with it your concurrency, requests-per-minute, tokens-per-minute and tokens-per-day limits. Vouchers don't count toward that total. Web search is a separate tool billed per call, with its own queries-per-second quota that doesn't eat into your chat limits.
The Batch API runs kimi-k3 at 60% of the standard price for jobs that don't need real-time answers — a straightforward saving for evaluations, back-filling and document processing. K3's full weights are published on Hugging Face (moonshotai/Kimi-K3, a 2.8T-parameter model), so self-hosting is possible if you have GPU capacity on that scale. Perplexity's Agent API also lists kimi-k3 at the same $3.00 / $15.00 rates, which helps if you already use its search tools.
Moonshot has since launched Kimi K3.1 Pro, a version of K3 with much more post-training and the same 1M-token window, plus a K3.1 Pro Highspeed variant that Moonshot says runs 2-3× faster. K3 is now listed as the previous-generation flagship. K3.1 Pro's prices are on Moonshot's pricing page but aren't tracked in the tables here yet; it gets the same 60% Batch rate, and Moonshot says its full weights will be released by Nov 4. K2-series models closed to newly registered users on October 31, 2026 and are scheduled to retire by the end of December 2026, so new projects should start on K3 or K3.1 Pro.
Kimi K3 is $3.00 per 1M input tokens (cache miss), $0.30 per 1M cached input tokens and $15.00 per 1M output tokens. Cache writes are billed separately, at the input rate for a 5-minute TTL or double for 1 hour.
Yes. Batch inference for kimi-k3 is billed at 60% of the standard price, for workloads that don't need real-time responses.
Yes. Moonshot has released the full K3 weights, published on Hugging Face as moonshotai/Kimi-K3, so you can run it yourself or through third-party hosts instead of the Kimi API.
Limits rise with your account's cumulative top-up amount, which determines your tier. Vouchers don't count toward it.
On list price it undercuts Claude Opus 5.5 ($4.00 / $20.00) and GPT-6 Astra ($10.00 / $50.00), but costs more than Gemini 4 Argon ($2.00 / $10.00) and Qwen3.8-Max ($2.00 / $6.00) per 1M tokens.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.