Last updated: 2026-11-02
What changed
- · Release Moonshot AI releases Kimi K3, a 2.8-trillion-parameter open model source
Kimi is Moonshot AI's assistant and API, built on the open-weight Kimi K3 models. The newest model in that family reads up to 1,048,576 tokens of context and ships with a Highspeed variant that Moonshot says runs two to three times faster. The app layers Deep Research, document and slide work, and parallel Swarm agents on top.
About Kimi
Kimi is the consumer assistant and developer platform from Moonshot AI, a Beijing lab founded in 2023. It began as a long-context chatbot and now runs on Moonshot's own open-weight models: Kimi K3, which reached the API in July 2026 and is billed by Moonshot as the first open-source model in the 3-trillion-parameter class, and Kimi K3.1 Pro, a post-trained K3 released in October 2026.
The app is now organised around agents rather than plain chat. Paid members get Deep Research, document, spreadsheet and slide work, website deployment with a database, scheduled tasks that push their results to you, a widget dashboard, plugins for financial and economic databases, and Swarm, which splits one job into parallel subtasks. Kimi Code comes with every paid plan, and Kimi Claw runs on the web, Android and PC. Moonshot labels Swarm, scheduled tasks and plugins as research previews, so expect them to keep changing.
Developers buy the same models through the Kimi API Platform, metered per token with no subscription. Rate limits rise with cumulative spend, the platform added a Web Search API in September 2026 and Batch support for the K3 models in October, and the docs carry setup guides for Codex, OpenCode and Hermes Agent. Moonshot retired kimi-k2.5 and the moonshot-v1 line in August 2026 and plans to retire every K2 model by the end of December, so older integrations need a migration plan.
Who might prefer something else: ChatGPT has the larger app ecosystem and more mature admin controls for big companies. Claude suits careful long-document writing. A team that must keep data outside China will usually start with one of those two. For a cheap open-weight API the closer like-for-like choice is DeepSeek. Shops already running on Alibaba Cloud may prefer Qwen. Kimi quotes paid agent credits relative to an allowance it never states, so heavy users cannot work out their ceiling before paying. For a head-to-head read, see ChatGPT vs Kimi.
Pricing
Consumer plans on kimi.com, in US dollars: Free $0; Plus $19 a month or $180 a year; Pro $39 a month or $372 a year; Max $99 a month or $948 a year; Ultra $199 a month or $1,908 a year. Business Premium is $600 per seat per year, billed annually only, with a two-seat minimum; 150 or more seats is a custom quote. API, per million tokens on the Kimi API Platform: kimi-k3.1-pro $2.00 input, $10.00 output and $0.20 cached input; kimi-k3.1-pro-highspeed $4.00, $20.00 and $0.20; kimi-k3 $3.00, $15.00 and $0.30.
On the K3 models cache writes are billed by TTL (5 minutes at the input rate, 1 hour at twice it), Batch jobs cost 60% of the standard rate, and prices exclude tax. The older kimi-k2.7-code and kimi-k2.6 endpoints ($0.95 in, $4.00 out) closed to new accounts on 31 October 2026. Read on 2 November 2026 from kimi.com's membership page and platform.kimi.ai's pricing page.
For another agent-first assistant's plan ladder, see Manus.
| Tier | Monthly price | What it includes |
|---|---|---|
| Free | Free | Chat on kimi.com and the mobile apps, 1 concurrent agent task, 2 scheduled tasks, 2 projects, 500MB project storage |
| Plus | $19/mo | Annual option $180/yr. Paid agent credits, Deep Research, documents, sheets and slides, website deployment, Kimi Code, Swarm with 2 subtasks, 10 scheduled tasks |
| Pro | $39/mo | Annual option $372/yr. Twice the Plus agent credits, K3.1 Pro Highspeed, Goal mode, one-click Kimi Claw, Swarm with 4 subtasks, 15 scheduled tasks |
| Max | $99/mo | Annual option $948/yr. Five times the Plus agent credits, K3 chat with up to 1M tokens of context, 4 concurrent agent tasks, 100 projects, 50GB storage |
| Ultra | $199/mo | Annual option $1,908/yr. Ten times the Plus agent credits, everything in Max, 25 scheduled tasks |
| Business Premium | Custom | $600 per seat per year, annual billing only, minimum 2 seats; no training on business data by default, Kimi Work, dedicated support |
| Business Custom | Custom | 150 or more seats, quote from Moonshot sales |
| API: kimi-k3.1-pro | Custom | Pay as you go, no subscription |
| API: kimi-k3.1-pro-highspeed | Custom | Pay as you go, faster variant of K3.1 Pro |
| API: kimi-k3 | Custom | Pay as you go, previous flagship |
Feature Comparison by Tier
| Feature | Max | Pro | Free | Plus | Ultra |
|---|---|---|---|---|---|
| Concurrent agent tasks | 4 | 2 | 1 | 2 | 4 |
| Agent credits | 5x Plus | 2x Plus | Not published | Base paid allowance | 10x Plus |
| Swarm subtasks | 8 | 4 | — | 2 | 8 |
| Scheduled tasks | 20 | 15 | 2 | 10 | 25 |
| Projects | 100 | 20 | 2 | 20 | 100 |
| Project storage | 50GB | 20GB | 500MB | 20GB | 50GB |
| K3.1 Pro Highspeed | ✓ | ✓ | — | — | ✓ |
| 1M-token K3 chat | ✓ | — | — | — | ✓ |
Key Features
- Newest flagship model: Released in October 2026 as a post-trained Kimi K3, with a Highspeed version that Moonshot rates at two to three times the speed; full weights are promised by 4 November.
- Open-weight K3 family: Kimi K3 reached the API in July 2026 with its weights published, and Moonshot calls it the first open-source model at 2.8 trillion parameters.
- Swarm parallel agents: Splits a job into parallel subtasks: two at a time on Plus, four on Pro and eight on Max and Ultra, per the plan comparison table.
- Scheduled tasks and dashboard: Tasks run on a timer and push results to a widget dashboard, from 2 scheduled tasks on Free up to 25 on Ultra.
- Documents, research and websites: Paid plans add Deep Research, document, spreadsheet and slide editing, and website deployment backed by a database, with 15+ plugin types for financial and economic data.
- Kimi Code and Kimi Claw: Kimi Code is included from Plus, and Pro adds one-click deployment of Kimi Claw, which runs on web, Android and PC with up to 10 group chats.
- Metered developer API: The Kimi API Platform bills per token with no subscription, runs Batch jobs at 60% of the standard rate and added a Web Search API in September 2026.
Pros
- Agent features arrive early in the plan ladder: scheduled tasks, the dashboard and Swarm all start at Plus rather than only on the top plan.
- Open weights for the K3 family mean a team can trial Kimi in the app and later self-host the same model family, which closed assistants do not allow.
- The API pricing page spells out input, output, cache-hit, cache-write and batch rates per model, so costs can be modelled before you commit.
Cons
- Moonshot warns that requests to its newest Pro model may queue or fail while launch demand is high.
- Paid plans describe usage as multiples of agent credits without publishing the base allowance, so you cannot size a plan in advance.
- Every K2 API model is due to retire by the end of December 2026, so integrations built on them must migrate to the K3 series.
- Moonshot AI is based in Beijing; teams with data-residency rules outside China should read the Kimi OpenPlatform privacy policy before sending sensitive data.
- Business Premium is sold on annual billing only, with no monthly option and no published discounts.
Product Information
- Cloud
- Yes
- Self-Hosted
- Yes
- On-Premise
- No
- Languages
- English, Chinese
- Training
- Kimi API Platform documentation, Model quickstart guides, Agent setup guides (Codex, OpenCode, Hermes Agent)
Data Handling
- Training-data policy
- The Kimi API Platform says it does not use enterprise customer data to train models by default, and offers Zero Data Retention to enterprise customers on request. Kimi Business plans state no model training on business data by default. The pricing page does not state the policy for Free and individual plans.
Frequently Asked Questions
What does Kimi actually cost?
On kimi.com the Free plan costs nothing, Plus is $19 a month, Pro $39, Max $99 and Ultra $199, and paying yearly cuts each by about a fifth. Business Premium is $600 per seat per year. The API is billed separately by usage with no subscription: the flagship Pro model costs $2.00 per million input tokens and $10.00 per million output tokens.
Can you use Kimi without paying?
Yes. The Free plan allows one agent task at a time, two scheduled tasks and two projects with 500MB of project storage, but it has no Swarm, no Highspeed model and no 1M-token chat. Moonshot does not publish a free message count. The API has no free plan; its flagship models open up after a first top-up of at least $1.
What are Kimi's closest competitors?
DeepSeek is the nearest match if you want a low-cost open-weight model family from a Chinese lab with a simpler app. Qwen suits teams already building on Alibaba Cloud. Manus fits people who want an autonomous agent workspace rather than a chat app with agents added, and ChatGPT stays the default for breadth of integrations.
How does Kimi compare to ChatGPT in 2026?
Kimi's K3 models ship as open weights a team can run on its own servers, and its API rates are published model by model; ChatGPT's models are closed. ChatGPT has the larger plug-in ecosystem, more mature enterprise admin and data handling outside China. Pick Kimi for open weights and agent features on mid-priced plans, and ChatGPT if compliance paperwork decides the purchase.
How do you set up Kimi?
Open kimi.com or the iOS or Android app and sign in to start on the Free plan. To build with it, create a key on platform.kimi.ai, top up your balance so the K3 models become available, and call the flagship by the model ID on the pricing page. The docs have step-by-step guides for Codex, OpenCode and Hermes Agent, and Kimi Code CLI works with the same key.
Top Alternatives
- ChatGPT: Kimi ships open-weight models and a metered API with published per-model rates; ChatGPT counters with a larger app ecosystem and more mature enterprise admin.
- DeepSeek: Both come from Chinese labs with open-weight models; Kimi wraps its K3 family in a fuller agent app with Swarm, scheduled tasks and Deep Research.
- Qwen: Qwen fits teams already on Alibaba Cloud; Kimi leads on its agent app and the long-context K3 models.
Hokai guides covering Kimi
- Kimi vs ChatGPT: Is Kimi K3 Better, and for Whom?: Kimi or ChatGPT? Kimi K3 is open-weights, with plans from $19 a month. ChatGPT Plus is $20 with GPT-6.1 Sol, which scores higher and costs less per token.
Also mentioned in
- ChatGPT Alternatives in 2026: Seven Reasons to Leave and the Pick for Each: ChatGPT alternatives sorted by why you are leaving: writing, Google, sources, privacy, Microsoft, API cost. One pick each, with prices checked 5 Nov 2026.
- Z.ai (GLM) Alternatives in 2026: Coding Plans, APIs and Older GLM Models: Z.ai and GLM alternatives in 2026: Kimi Code and MiniMax's M Plan against the GLM Coding Plan, DeepSeek and MiniMax for the API, and what replaces old GLM.
- The Best Android AI Assistant in 2026, Free and Paid, Now That Google Assistant Is Gone: Gemini is the default Android assistant since Google Assistant's 4 Sep 2026 shutdown. ChatGPT, Perplexity, Copilot, Meta AI and Claude compared; all start free.
- What Together AI Is Actually For, Now IBM Is Betting $240 Million on It: IBM is spending $240M on Together AI's inference cluster and Together raised $800M in July. Here's what changed for buyers this week, and what did not.