Moonshot AI launched Kimi K3.1 Pro on October 23, 2026 at $2.00 per million input tokens and $10.00 per million output tokens on its official API, with cached input at $0.20. Kimi K3, the model it succeeds, stays on the price list at $3.00 and $15.00. The newer and, by Moonshot's account, better model costs a third less than the one before it.
That is the number to hold onto. Moonshot did not cut K3's price in the three months it has been live. It released a stronger model at a lower price point and left the old one where it was. This page has the official rates, the Highspeed tier, how K3.1 Pro compares with the other open-weight flagships, who sells it today, and the workload math. For the model it replaces, see Kimi K3 API pricing.
Kimi K3.1 Pro official API pricing
Rates from Moonshot's pricing page, USD per 1M tokens:
| Token type | kimi-k3.1-pro | kimi-k3.1-pro-highspeed |
|---|---|---|
| Input | $2.00 | $4.00 |
| Cached input | $0.20 | $0.20 |
| Output | $10.00 | $20.00 |
| Cache write, 5 minute TTL | $2.00 | $4.00 |
| Cache write, 1 hour TTL | $4.00 | $8.00 |
| Context window | 1,048,576 tokens | 1,048,576 tokens |
Pricing is flat across the full context window with no long-context surcharge. Cached input is billed at a 90 percent discount to the cache-miss rate. Cache writes are billed separately by how long the entry lives: the 5 minute tier costs the same as normal input, the 1 hour tier costs double. Caching is on by default with a 5 minute TTL.
Three things to know about how the model bills. Thinking is always on, with a reasoning_effort setting of low, high, or max, and max is the default, so output token counts will run higher than a non-reasoning model at the same task unless you turn it down. Maximum output defaults to 131,072 tokens and can be raised to the full 1,048,576. And the Highspeed tier is the same model at double the price for what Moonshot says is two to three times the speed; on OpenRouter the standard endpoint shows 30 tokens per second and Highspeed 56.
What changed from Kimi K3
Moonshot describes K3.1 Pro as a post-trained version of Kimi K3 with substantially larger post-training compute, in the same 3T-class architecture. The gains it reports are in the places an institutional buyer would look: a claim-level hallucination rate down from 23.35 percent on K3 to 6.85 percent, and internal finance benchmarks up 7.2 points on financial modeling and 11.6 points on financial research. Those are Moonshot's own numbers and have not been independently reproduced.
The price moved the other way. K3 launched at $3.00 and $15.00 in July and has not changed. K3.1 Pro arrives at $2.00 and $10.00, a 33 percent lower rate on every billable token type, and K3 stays on the price list at the old rate. A buyer on K3 today pays a 50 percent premium over the model Moonshot says is better.
Moonshot is also clearing the back catalog. From October 31, 2026 the K2 series is closed to new registrations, and Moonshot plans to retire all K2 models by the end of December 2026. Anyone still on K2.6 or K2.7 Code has two months to move.
How K3.1 Pro compares with other open-weight flagships
Official rates, USD per 1M tokens:
| Model | Input | Output | Source |
|---|---|---|---|
| Kimi K3.1 Pro | $2.00 | $10.00 | Moonshot |
| Kimi K3 | $3.00 | $15.00 | Moonshot |
| Mistral Large 4 | $1.36 | $4.18 | Mistral |
| DeepSeek V4 Pro, peak | $1.32 | $3.96 | DeepSeek |
| DeepSeek V4 Pro, off-peak | $0.66 | $1.98 | DeepSeek |
| Meta Muse Spark Open, hosted | $1.25 | $4.25 | Baseten, Together, Fireworks |
Even after the cut, K3.1 Pro is the most expensive open-weight flagship on an official API. Its output rate is 2.4 times Mistral Large 4, 2.5 times DeepSeek V4 Pro at peak, and 5 times V4 Pro off-peak. Moonshot is pricing it as a reasoning model that competes with closed frontier models, not with the open-weight field on cost. Whether that holds depends on what happens when the weights are out.
Who sells Kimi K3.1 Pro today
One company. OpenRouter lists two endpoints for K3.1 Pro, both Moonshot's: the standard tier at $2.00 and $10.00 and the Highspeed tier at $4.00 and $20.00. No third party can host the model until the weights are released, and Moonshot says that happens by November 4, 2026. The license has not been published.
The K3 precedent says what comes next. A month after K3's weights went up in July, eleven hosts had listed it within 1.23x of each other. Three months on there are 20 providers across 24 listings, and the same model sells from $0.10 to $4.50 per million input tokens. If K3.1 Pro follows that curve, the official $2.00 rate will be the ceiling of the third-party market within weeks of November 4, not the floor. Why token prices differ explains where that spread comes from.
What a real workload pays
A production assistant doing 10 million input tokens and 1 million output tokens a day, uncached, at official list rates on October 25, 2026. The K3 third-party row uses the cheapest host on OpenRouter's standard routing.
| Route | Daily | Annual |
|---|---|---|
| Kimi K3.1 Pro, official | $30.00 | $10,950 |
| Kimi K3.1 Pro Highspeed, official | $60.00 | $21,900 |
| Kimi K3, official | $45.00 | $16,425 |
| Kimi K3, cheapest third party (Relace) | $5.00 | $1,825 |
| Mistral Large 4, official | $17.78 | $6,490 |
| DeepSeek V4 Pro, official peak | $17.16 | $6,263 |
| DeepSeek V4 Pro, official off-peak | $8.58 | $3,132 |
Three things fall out of the table.
Moving from K3 to K3.1 Pro on the official API saves $15 a day on this workload, $5,475 a year, for a model Moonshot says is better. There is no reason to stay on K3 at the official rate once your evals pass.
The cheapest K3 host is still a sixth of the K3.1 Pro official price. Relace serves the old model at $0.10 input and $4.00 output, which runs the workload for $5 a day against $30. A buyer whose task was fine on K3 and who can tolerate a smaller host pays less for the older model than Moonshot charges for the newer one. That gap closes only when K3.1 Pro weights reach third parties.
Cache changes the number more than the model choice does. At a 60 percent cache hit rate, the K3.1 Pro official route drops to $19.20 a day before cache write fees, because 6 million of the 10 million input tokens bill at $0.20 instead of $2.00. The 5 minute write tier costs the same as input, so a prefix written once and read many times pays for itself on the second hit.
Which Kimi model to use
Running K3 on the official API: test K3.1 Pro on your workload. The price is a third lower and Moonshot claims the quality is higher, so the only question is whether it passes your evals. Change the model name and the bill drops.
Running K3 on a third-party host: stay put until November. The cheapest K3 hosts are far below the K3.1 Pro official rate, and the newer model has no third-party market yet.
Latency-sensitive: Highspeed doubles the price for two to three times the throughput. At $60 a day on this workload it costs more than K3 did. Pay it only where tokens per second has a value you can measure.
Still on K2: move now. Registrations close October 31 and the models retire by the end of December.
For the model-by-model view across DeepSeek, GLM, Qwen, and Kimi, see the LLM API pricing comparison. To track official and third-party rates as they move, see the Mercatus Token Index.
Where this leaves a Kimi buyer
Moonshot's pricing history is the useful signal. K3's rate has not moved since July. The company did not compete with the third-party market that undercut it; it shipped a better model at a lower price and left K3 in place. Better models create competition, and this is what that looks like on a rate card: the price of the capability fell by a third while the price of the old product stayed exactly where it was.
What the buyer still lacks is a way to hold a rate. Nothing stops Moonshot from repricing K3.1 Pro once the weights are out and the third-party market forms, in either direction. How to lock in token prices covers what a buyer with predictable volume can do about that, and how a token exchange works covers how a single market price gets set once more than one seller is quoting the same model.
Frequently asked questions
How much does the Kimi K3.1 Pro API cost?
$2.00 per million input tokens, $0.20 per million cached input tokens, and $10.00 per million output tokens on Moonshot's official API. Cache writes are $2.00 per million for a 5 minute TTL and $4.00 for a 1 hour TTL. The Highspeed tier is $4.00 input and $20.00 output with the same $0.20 cached rate.
Is Kimi K3.1 Pro cheaper than Kimi K3?
Yes, by a third on every token type. K3 is $3.00 input, $0.30 cached, and $15.00 output. K3.1 Pro is $2.00, $0.20, and $10.00. K3 stays available at its old price.
What is the model name for Kimi K3.1 Pro in the API?
kimi-k3.1-pro for the standard tier and kimi-k3.1-pro-highspeed for the faster tier. Both have a 1,048,576 token context window.
Is Kimi K3.1 Pro open weight?
Moonshot says the full weights will be released by November 4, 2026. As of October 25, 2026 they are not available and the license has not been published.
Who hosts Kimi K3.1 Pro?
Only Moonshot. OpenRouter lists two Moonshot endpoints, standard and Highspeed. Third-party hosting starts after the weights are released.
When are Kimi K2 models retired?
From October 31, 2026 K2 series models are closed to newly registered users. Moonshot plans to retire all K2 models by the end of December 2026.
Methodology
K3.1 Pro, Highspeed, and K3 rates, cache terms, context window, and the K2 retirement schedule are from Moonshot's Kimi API pricing page. Model names, reasoning settings, output limits, the weights date, and the hallucination and finance benchmark figures are from Moonshot's K3.1 Pro quickstart documentation. Launch date, host count, and throughput are from OpenRouter's model page. K3 third-party prices are from the Mercatus Kimi K3 pricing page, OpenRouter standard routing as of October 8, 2026. Mistral and DeepSeek rates are from those vendors' pricing pages as documented on their Mercatus pages. Workload examples use stated token volumes at 365 days; recompute with your own mix. Last verified: 2026-10-25.