--- title: "Kimi K3.1 Pro Pricing: API Cost, Highspeed, Cache Rates, and How It Compares" slug: kimi-k3-1-pro-pricing-api-cost description: "Kimi K3.1 Pro costs $2 per million input tokens, $0.20 cached, and $10 per million output on Moonshot's API, a third less than Kimi K3. The full rate card including Highspeed and cache writes, what a real workload costs with and without caching, Kimi Code plan prices, and how it stacks up against the other frontier and open models." author: "Yotta Labs" date: 2026-10-27 categories: ["Inference"] canonical: https://www.yottalabs.ai/post/kimi-k3-1-pro-pricing-api-cost --- # Kimi K3.1 Pro Pricing: API Cost, Highspeed, Cache Rates, and How It Compares ![](https://cdn.sanity.io/images/wy75wyma/production/7fdb195fba8f888a23bf4e3b6260b960382c46f0-1200x627.png) *Every rate Moonshot charges for Kimi K3.1 Pro, what caching does to the bill, and what the same workload costs on the models it competes with.* Moonshot released [Kimi K3.1 Pro](https://www.yottalabs.ai/post/kimi-k3-1-pro-specs-price-open-weights-benchmarks) on October 23, 2026, and priced it below the model it replaces. That is unusual for a flagship upgrade, and it makes the pricing worth a post of its own. This one covers the complete rate card from Moonshot's API pricing page, including the cache-write charges most summaries skip, the Highspeed variant, the Kimi Code subscription plans, and a side-by-side bill against DeepSeek V4 Pro, GLM 5.3, Qwen3.8-Max, Kimi K3, Muse Wonder, and GPT-6 Astra. ## The short version - **Kimi K3.1 Pro:** $2.00 per million input tokens, $0.20 per million cached input, $10.00 per million output. - **Kimi K3.1 Pro Highspeed:** $4.00 input, $0.20 cached input, $20.00 output. Same model, two to three times faster per Moonshot. - **Kimi K3:** $3.00 / $0.30 / $15.00. K3.1 Pro is a third cheaper on every rate. - **Cache writes** are billed separately: $2.00 per million for a five-minute cache and $4.00 for a one-hour cache on K3.1 Pro. - **Kimi Code plans:** $19, $39, $99, and $199 a month, less on annual billing. - **Against the field:** half of Muse Wonder's bill, a fifth of GPT-6 Astra's, and about two and a half times DeepSeek V4 Pro on Yotta AI Gateway. ## Kimi K3.1 Pro API pricing Per million tokens, from Moonshot's API pricing page. Prices exclude tax. Three things to read off the table. The output rate is where the money goes. At $10 per million, output is five times the input rate, so a workload that generates long answers or long agent traces pays mostly for output, and caching doesn't touch that side. Highspeed is exactly double on input and output but the same $0.20 on cached input. If your prompts are mostly cached context, the Highspeed premium is smaller than the headline doubling suggests. K3 stays on the price list at the old rates. There's no reason to pay them: K3.1 Pro is the same architecture, Moonshot's benchmarks put it well ahead, and it costs less. ## How caching changes the bill Moonshot's context caching is automatic. Repeated request prefixes (system prompts, tool definitions, documents, earlier turns) are cached and billed at the cache-hit rate on later requests. Two things are new with the K3 series: the cache write is a separate charge, and you choose a time-to-live of five minutes or one hour depending on how far apart your calls are. What that means in practice: the first time a prefix is sent you pay the input rate plus the write rate, so $4.00 per million on K3.1 Pro with the five-minute TTL. Every later request that hits the cache within the TTL pays $0.20 per million for that prefix. For an agent that resends the same 50K-token context on every turn, the break-even is the second call. Here's the same daily workload, 10 million input tokens and 2 million output, with and without caching. The cached case assumes 80% of input tokens are cache hits and every miss is written with the five-minute TTL, which is the simplest assumption and slightly overstates write costs. Caching takes about a quarter off a prompt-heavy workload. It can't do more than that here because output is $20 of the $40 and output doesn't cache. On a chat workload with short prompts and long answers, caching barely moves the number. ## Kimi Code plans Kimi Code is Moonshot's coding agent (CLI, VS Code extension, desktop app), and K3.1 Pro is selectable in it with the /model command per Moonshot's launch post. The plans on Moonshot's Kimi Code page as of October 27: The page says K3-class models are available on Plus and above, and describes the tiers by usage quota and concurrency rather than by published request counts. Moonshot hasn't published the quotas behind each tier, so treat the plan as a seat for a person using Kimi Code, the same way Muse Code and the ChatGPT Pro plans work on the closed side. For an application, the API is the route. ## What the same workload costs elsewhere 10 million input tokens and 2 million output a day, no caching, 30 days. Gateway rates are Yotta AI Gateway's flat per-token prices. Two reads. Against the closed frontier, K3.1 Pro is cheap. Half of Muse Wonder and a fifth of GPT-6 Astra, for a model Moonshot puts within a point of Claude Opus 5.5 on Terminal-Bench 4.0. The [GPT-6 Astra pricing](https://www.yottalabs.ai/post/gpt-6-astra-pricing-api-cost-2026) and [Muse Wonder pricing](https://www.yottalabs.ai/post/meta-muse-wonder-pricing-api-cost-muse-code-plans) posts cover those two in detail. Against the open flagships, it's the expensive one. DeepSeek V4 Pro, GLM 5.3, and Qwen3.8-Max on Yotta AI Gateway run at 40% to 60% of K3.1 Pro's bill. The difference is mostly the output rate: $10 against $2.97 to $4.50. The question for a team is whether K3.1 Pro's extra capability shows up in the output enough to pay for it on that traffic, and the only way to know is to run the same prompts through both. ## Where the price goes when the weights land Moonshot says the full K3.1 Pro weights will be released by November 4, 2026. For most teams that won't change the bill, because the model is 2.8 trillion parameters with 104B active and needs a multi-node cluster with 2 TB or more of GPU memory. The [Kimi K3 hardware requirements](https://www.yottalabs.ai/post/kimi-k3-hardware-requirements-gpu-memory-2026) post has the math, and it applies to K3.1 Pro unchanged. What open weights do change is the hosted market. Kimi K3 is listed by more than two dozen hosts on OpenRouter today, most at Moonshot's $3 / $15 and several below it, down to $1.20 / $5.50. Expect the same spread for K3.1 Pro once the weights are out, with Moonshot's $2 / $10 as the reference rate rather than the floor. We'll update this post with hosted rates as they appear. ## How to use it today Call kimi-k3.1-pro on Moonshot's API. It's OpenAI-compatible, so an existing client switches with a base URL and model string. Use kimi-k3.1-pro-highspeed only where latency is worth double the token price. For the comparison set, [Yotta AI Gateway](https://www.yottalabs.ai/ai-gateway) puts Kimi K3, DeepSeek V4 Pro, GLM 5.3, and Qwen3.8-Max behind one OpenAI-compatible key with flat per-token pricing. Running K3.1 Pro against them is a model-string change in the same client, and the [best OpenAI API alternatives](https://www.yottalabs.ai/post/best-openai-api-alternatives-in-2026-free-open-source-and-multi-model-options) post covers that routing pattern. ## Frequently asked questions **How much does Kimi K3.1 Pro cost?** $2.00 per million input tokens, $0.20 per million cached input, and $10.00 per million output on Moonshot's API. Cache writes are $2.00 per million for a five-minute cache and $4.00 for one hour. **How much does Kimi K3.1 Pro Highspeed cost?** $4.00 per million input, $0.20 cached input, $20.00 per million output. Cache writes are $4.00 and $8.00 per million for the five-minute and one-hour TTLs. **Is Kimi K3.1 Pro cheaper than Kimi K3?** Yes, by a third on every rate. K3 is $3.00 / $0.30 / $15.00. **Is Kimi K3.1 Pro cheaper than GPT-6 Astra?** Yes, by 80%: $2 against $10 per million input, $10 against $50 per million output. **Is Kimi K3.1 Pro cheaper than Muse Wonder?** Yes, by half: Muse Wonder is $4 / $20. **How does caching work on Kimi K3.1 Pro?** Repeated request prefixes are cached automatically. The first send pays the input rate plus a cache write, and later hits within the TTL pay $0.20 per million. You pick a five-minute or one-hour TTL. **What does Kimi Code cost?** $19, $39, $99, or $199 a month on monthly billing, or $15, $31, $79, and $159 on annual billing, per Moonshot's Kimi Code page. **Is there a free tier for Kimi K3.1 Pro?** Moonshot's pricing page lists per-token rates only. Check the Kimi API platform for any current credits. **Is Kimi K3.1 Pro on Yotta?** Kimi K3 is live on Yotta AI Gateway at $3.00 per million input and $15.00 per million output, alongside DeepSeek V4 Pro, GLM 5.3, and Qwen3.8-Max. **Will Kimi K3.1 Pro get cheaper?** Probably, once the weights are out on November 4 and third-party hosts list it, as they did with K3. Moonshot's rate is the reference price today. ## Bottom line Kimi K3.1 Pro is priced at $2 in and $10 out, a third under K3, half of Muse Wonder, and a fifth of GPT-6 Astra. Caching takes about a quarter off a prompt-heavy agent workload and nothing off output. Highspeed doubles the token price for two to three times the speed. For an application, decide by cost per result. Run it against the open flagships on [Yotta AI Gateway](https://www.yottalabs.ai/ai-gateway) with the same prompts, and let the output decide whether the extra $5 to $7 per million output tokens is buying anything.