Use Kimi K3 in Your IDE Without a Subscription
Bring Kimi K3 into VS Code, Claude Code, Zed, JetBrains, OpenCode or Neovim. Start small with no monthly IDE plan or long-term commitment.
What to do
- Open Novita with the affiliate link and create or sign in to your account.
- Find moonshotai/kimi-k3 and confirm the live token price before adding funds.
- Create an API key in Novita and copy it to a secure password manager.
- Choose your coding client: VS Code or JetBrains with Continue, Claude Code, Zed, Cline, Roo Code, Neovim or another compatible tool.
- Enter Novita's compatible endpoint, the model ID moonshotai/kimi-k3 and your API key using the matching instructions below.
- Test Kimi K3 with one small, non-sensitive project before authorizing wider file or terminal access.
- Check Novita usage and cost after the first few prompts before increasing use.
This approved affiliate link tracks the referral but does not apply a coupon, exclusive discount or guaranteed credit. Check the live model price and any account voucher before adding funds.
I may receive a reward if you use this link or code.
Kimi K3 and Novita details checked
This is a current coding setup rather than a coupon. These are the practical details that matter before adding a payment method.
- Model
- moonshotai/kimi-k3Novita's exact model ID for the OpenAI-compatible API.
- Release status
- Current flagship Kimi modelKimi describes K3 as its flagship for long-horizon coding and knowledge work.
- Context
- 1,048,576 tokensNovita and Kimi both list a context window of up to 1M tokens.
- Current API price
- $3 input / $15 output per 1M tokensCache reads are listed at $0.30 per 1M tokens. Prices are in USD and can change.
- Visitor promotion
- No exclusive discountThe affiliate link tracks eligible spending; it does not promise a coupon or fixed free credit.
- Cursor comparison
- Bring K3 to your preferred coding clientAgent and editing features depend on the client; custom API access is not an exact clone of Cursor.
There is no verified customer promo code or fixed signup credit attached to this link. Novita's older Give $10, Earn $10 referral campaign is marked ended. Kimi K3 usage is billed by token and can cost more or less than a flat coding subscription depending on use. Compatibility and available features vary by client: agent tools, file edits, terminal access, MCP, image input, autocomplete and tab prediction are not identical across VS Code extensions, Claude Code, Zed, JetBrains and Neovim plugins. Prices, models, limits and free account vouchers can change. Code and prompts are processed through Novita and its model infrastructure, so review current data terms before sending private or restricted code.
- Novita official Kimi K3 model, price and API details
- Novita current model API pricing
- Novita official OpenAI-compatible LLM API guide
- Novita API key management guide
- Novita official Continue setup for VS Code and JetBrains
- Novita official OpenCode custom-provider setup
- Novita official Codex custom-provider requirements
- Novita official Claude Code integration guide
- Novita official Anthropic-compatible API guide
- Novita official CodeCompanion setup for Neovim
- Cline official OpenAI-compatible provider guide
- Kimi official model-selection guide for K3
- Kimi official K3 plans and API pricing guide
- Cursor official bring-your-own API key documentation
- Zed official custom OpenAI-compatible provider guide
- Zed official Edit Prediction guide
- Anthropic position on third-party Claude Code gateways
- Novita public affiliate program terms
- Novita ended Give $10, Earn $10 referral page
Offer details checked against these provider pages on Jul 27, 2026.
What the Novita Kimi K3 API offer actually gives you
Kimi K3 is current and relevant for coding. Kimi calls it the flagship model for long-horizon coding and end-to-end knowledge work, and Novita exposes it through serverless APIs that use common OpenAI-compatible and Anthropic-compatible formats. That lets developers bring K3 into coding tools instead of being limited to a browser chatbot.
This is a general Kimi K3 API offer, not a Zed product. The same Novita account and API key can be used with compatible clients such as Continue in VS Code or JetBrains, Claude Code through Novita's Anthropic-compatible endpoint, Zed through a custom OpenAI-compatible provider, Cline or Roo Code in VS Code, CodeCompanion in Neovim, and other tools that accept a custom endpoint, model ID and API key.
Novita supplies access to the model; the coding client supplies the interface and tools. Depending on the client, K3 may be able to explain a repository, plan changes, edit files, run commands, use tools, review code or work through a longer task. Those capabilities are not created by the affiliate link itself and should be confirmed in the specific client before relying on them.
This can replace a large part of a Cursor workflow, but calling Novita alone a complete Cursor replacement would be inaccurate. K3 supplies the reasoning and coding model while Continue, Claude Code, Zed, Cline, Roo Code or another client supplies the editor integration and agent features. Autocomplete and tab prediction may use a separate model or provider.
Novita charges for actual token use instead of selling this setup as a fixed monthly coding subscription. K3 was listed at $3 per million input tokens, $0.30 per million cached input tokens and $15 per million output tokens when checked. A short coding question can cost very little, while repeated whole-repository prompts and long reasoning outputs can add up quickly.
The 1M-token context window is a ceiling, not a reason to send an entire repository every time. Give the agent only the files and context it needs, start with reasoning effort set to high rather than max, review the usage dashboard, and keep automatic top-ups off until you understand your normal cost.
The affiliate link on this page is useful for attribution, not for a secret customer deal. Referral Quest may receive 10% of eligible referred spending for Novita's stated 180-day commission period. Your Novita price is not increased by the link, but no exclusive coupon or guaranteed credit is attached to it.
If your priority is the newest Kimi model and very long context, choose K3. If your priority is minimizing API cost, compare other coding models in Novita's live catalogue before funding heavily. The best client and model combination depends on code quality, agent features, latency, context size, privacy requirements and how many tokens the tool sends.
Try Kimi K3 without a subscription
Metered pricing, no monthly fee, and it works in the editor you already use.
Affiliate link - no exclusive discount attached.What is the cheapest way to add Kimi K3 to an IDE?
For light or irregular coding use, the cheapest practical route may be a free coding client plus a pay-as-you-go Kimi K3 API. You avoid a fixed editor subscription and pay only for the tokens the client sends and receives. Heavy daily agent use can reverse that calculation, so compare your real monthly API total with a flat subscription.
A low-cost setup is Continue, Cline or Roo Code in VS Code; Continue in a JetBrains IDE; OpenCode in a terminal or desktop app; Zed with a custom provider; or CodeCompanion in Neovim. The client can be free, but K3 API use is not unlimited or free. When checked, K3 cost US$3 per million uncached input tokens, US$0.30 per million cached input tokens and US$15 per million output tokens.
The cheapest setup is not automatically the client with the lowest sticker price. Coding agents may resend system instructions, conversation history, repository context and tool results on every step. A task that looks like one request can contain several API calls. Start with a small balance and measure one real task before deciding whether API billing or a monthly coding plan is cheaper for you.
(uncached input tokens ÷ 1,000,000 × $3) + (cached input tokens ÷ 1,000,000 × $0.30) + (output tokens ÷ 1,000,000 × $15)| Example request | Uncached input | Output | Estimated cost |
|---|---|---|---|
| Small code question | 20,000 tokens | 2,000 tokens | US$0.09 |
| Focused feature or debugging task | 50,000 tokens | 10,000 tokens | US$0.30 |
| Larger repository task | 100,000 tokens | 20,000 tokens | US$0.60 |
| Very large first-pass context | 500,000 tokens | 50,000 tokens | US$2.25 |
| Same 500,000-token context if fully cache-read | 500,000 cached | 50,000 tokens | US$0.90 |
These are arithmetic examples, not quotes or maximums. An IDE agent can make multiple requests, caching is not guaranteed for every token, taxes and currency conversion may apply, and Novita or Kimi can change pricing.
Which IDE or coding agent should you use with Kimi K3?
Novita supplies the K3 API. Your chosen client supplies the editor interface, repository search, file editing, terminal tools, MCP support, approvals and autocomplete. Choose the client for the workflow you want rather than assuming every K3 integration behaves like Cursor.
| Coding client | Best for | Connection route | Important limitation |
|---|---|---|---|
| Continue for VS Code or JetBrains | Chat and agent workflows inside an existing IDE | Official Novita provider or OpenAI-compatible endpoint | Autocomplete can use a separate model and configuration |
| OpenCode | A flexible terminal or desktop coding agent | Custom provider at https://api.novita.ai/openai/v1 | Tool permissions and features depend on the OpenCode release |
| Zed | A fast editor with a custom K3 agent model | Custom OpenAI-compatible provider | Edit Prediction is separate from the K3 chat model |
| Cline or Roo Code | VS Code agents with file and terminal tools | OpenAI Compatible provider | Review approval settings before allowing automatic commands |
| CodeCompanion for Neovim | Neovim users who want chat and code actions | Novita adapter using its OpenAI-compatible API | Plugin configuration can change between releases |
| Claude Code | Terminal agent workflows through an Anthropic-style client | Novita Anthropic-compatible endpoint | Third-party non-Claude routing is not supported or guaranteed by Anthropic |
| Cursor | People already using Cursor's own agent and editor features | No clean, officially documented arbitrary Novita K3 route | Cursor BYOK and Tab features have provider and model restrictions |
| Codex CLI | Terminal coding through a custom Novita provider | Novita's documented Responses API configuration | Confirm K3 currently supports Novita's Responses route before choosing K3 |
Visual Studio and Visual Studio Code are different products. These VS Code extension instructions do not automatically apply to the full Visual Studio IDE.
Kimi K3 API vs Kimi Code subscription vs Cursor
These products overlap, but they are not the same purchase. This distinction matters when people search for a cheap Kimi K3 Cursor alternative or a way to use Kimi 3 in VS Code.
Novita and Kimi are separate providers. A Novita account does not require a Kimi.com membership, and a Kimi subscription is not used to pay a Novita API bill. Do not mix Kimi's first-party base URL or subscription token with Novita's endpoint and key.
Pay-as-you-go is attractive when you code occasionally because a quiet month can cost very little. A subscription can be better value when you run long agents every day, repeatedly attach a large repository or prefer a predictable bill. The only reliable comparison is your measured API usage over a normal month.
| Option | How you pay | What it includes | Best fit |
|---|---|---|---|
| Kimi K3 through Novita | Token-based API usage | Hosted model access that you connect to a compatible client | Light or variable use, custom workflows and developers who want provider choice |
| Kimi Code or Kimi membership | Kimi's current plan or quota system | Kimi's own coding or workspace experience | People who want Kimi's first-party interface and included plan allowances |
| Cursor subscription | Cursor's current plan price and limits | Cursor editor, agent, indexing, Tab and supported models | People who want an integrated product instead of assembling a client and API |
| Browser AI extension | Extension plan, credits or free limits | Web-page chat and browser assistance | Research and page questions, not a full repository-aware IDE agent |
How to make Kimi K3 cheaper for coding
Most savings come from controlling context and agent behaviour, not hunting for an unverified Kimi K3 coupon code.
- Start with only the files and folders needed for the task instead of attaching an entire repository.
- Open a fresh session when old conversation history is no longer useful.
- Use low reasoning effort for connection tests and simple questions; use high for serious coding before trying max.
- Ask for a concise plan or diff and cap output when the client exposes a maximum-output setting.
- Reuse stable context when the client and provider can take advantage of cache reads.
- Check Novita's usage screen after one small task, one medium task and one agent task before funding more.
- Keep automatic top-up off and set a small budget or low-balance alert while learning your usage.
- Use a less expensive model for routine summaries or formatting and reserve K3 for difficult coding and long-context work.
- Stop loops that repeat the same failed tool call; every retry can add input and output tokens.
- Never treat a temporary voucher as permanent pricing or an affiliate link as a guaranteed promo code.
Kimi's official pricing says a cache hit costs 90% less for input than a cache miss. Whether a particular IDE request receives that rate depends on the API and request pattern.
Kimi K3 IDE troubleshooting: 401, 404, missing model and high cost
Most failed setups come from mixing providers, using the wrong endpoint form or expecting a chat model to supply editor features that belong to the client.
| Problem | Likely cause | What to check |
|---|---|---|
| 401 unauthorized | Missing, invalid, revoked or badly pasted Novita key | Create a current Novita key, remove spaces and use the client's secret field |
| 404 endpoint error | The client added /v1 twice or needs it added once | Try the exact endpoint required by the matching guide rather than mixing examples |
| Model not found | Incorrect provider or model ID | Use the Novita account and exact ID moonshotai/kimi-k3 |
| Kimi K3 does not appear in a menu | The client uses a fixed model list | Look for Custom Provider, OpenAI Compatible or manual Model ID |
| Chat works but file edits do not | Client tools or permissions are not enabled | Test file, terminal and MCP access separately and review approval settings |
| No Cursor-style autocomplete | Tab prediction is a separate feature or model | Configure the client's autocomplete or edit-prediction provider separately |
| Unexpectedly high bill | Large context, long output or repeated agent steps | Inspect per-request usage, reduce context, cap output and stop retry loops |
| Claude Code tools behave differently | Third-party non-Claude routing is not guaranteed by Anthropic | Test the Novita route carefully or choose a client with native custom-provider support |
Do not paste an API key into a support post, screenshot or repository while troubleshooting. Revoke it immediately if it has been exposed.
Connect Kimi K3 to your coding tool
Choose the client you already use. Novita's OpenAI-compatible base is https://api.novita.ai/openai and its Anthropic-compatible base is https://api.novita.ai/anthropic. The exact path and available agent features depend on the client. Never commit an API key to a repository.
The three connection values every client needs
Create one Novita API key, then use the endpoint format your coding client supports.
- In Novita, open API Key Management and create a key. Copy it once and store it securely.
- For an OpenAI-compatible client, use https://api.novita.ai/openai/v1 as the base URL.
- For a client that accepts Anthropic-compatible providers, use https://api.novita.ai/anthropic as the base URL.
- Use the exact model ID moonshotai/kimi-k3.
- Paste the API key only into the client's provider or secret field, never into a project file.
- Start with one small prompt and confirm it appears in Novita's usage dashboard.
Some clients ask for a base URL without /v1 and add that path themselves. Follow the matching setup below; a duplicated or missing /v1 is a common cause of 404 errors.
VS Code or JetBrains: use Continue
Novita publishes an official Continue integration for both VS Code and JetBrains IDEs.
- Install Continue from the VS Code Marketplace or the JetBrains plugin marketplace.
- Open Continue and choose Add Chat Model.
- Select Novita AI as the provider.
- Enter the API key from your Novita account.
- Select Kimi K3 if it appears, or add moonshotai/kimi-k3 as the custom model ID.
- Ask K3 to explain one file before granting broader agent or command permissions.
Continue supplies the editor experience; Novita supplies the K3 model. Chat, autocomplete, agent and tool features can use separate models or settings inside Continue.
OpenCode: use Kimi K3 in a terminal or desktop coding agent
Novita publishes an official OpenCode integration for custom models. OpenCode is useful when you want a coding agent without changing your main editor.
- Install the current OpenCode command-line tool or desktop app from OpenCode's official site.
- Open OpenCode settings and find Providers or Custom Provider.
- Display name: Novita.
- Base URL: https://api.novita.ai/openai/v1.
- API key: use the key created in your Novita account.
- Model ID: moonshotai/kimi-k3.
- Save, select Kimi K3 via Novita and test a read-only project question before approving edits or commands.
Novita's OpenCode documentation confirms the custom-provider route and exact base URL. K3's model ID comes from Novita's Model Hub; context, price and capabilities remain model-specific.
Claude Code: use Novita's Anthropic-compatible endpoint
Novita publishes a Claude Code integration using its Anthropic-compatible API. This PowerShell setup applies only to the current terminal window.
- Install the current Claude Code release and open PowerShell in a small test project.
- Set ANTHROPIC_BASE_URL to Novita's Anthropic-compatible endpoint.
- Set ANTHROPIC_AUTH_TOKEN to your Novita API key.
- Set both model variables to moonshotai/kimi-k3 so background requests do not fall back to an unavailable Anthropic model.
- Run claude and confirm a simple read-only request works before approving edits, commands or MCP tools.
$env:ANTHROPIC_BASE_URL="https://api.novita.ai/anthropic"
$env:ANTHROPIC_AUTH_TOKEN="PASTE_YOUR_NOVITA_KEY"
$env:ANTHROPIC_MODEL="moonshotai/kimi-k3"
$env:ANTHROPIC_SMALL_FAST_MODEL="moonshotai/kimi-k3"
claudeNovita documents non-Claude models in Claude Code, but Anthropic does not support or guarantee third-party routing to non-Claude models. Claude Code updates can therefore affect tools or compatibility. Test the current versions before making it your main workflow.
Zed: add Novita as an OpenAI-compatible provider
Zed is one supported example, not the only way to use this offer.
- In Zed, open the command palette and run agent: open settings.
- Under LLM Providers, choose Add Provider and select an OpenAI-compatible provider.
- Provider name: Novita. API URL: https://api.novita.ai/openai/v1.
- Model ID: moonshotai/kimi-k3. Context window: 1048576.
- Set reasoning effort to high, enter the Novita key when prompted and save.
- Open Agent, select Kimi K3 via Novita and test one non-sensitive file.
K3 handles Agent and inline requests in this setup. Zed's separate Edit Prediction provider controls tab suggestions while you type.
Optional copyable Zed settings.json configuration
Add this inside your existing Zed settings object. If language_models already exists, merge the nested novita provider instead of duplicating the top-level key.
{
"language_models": {
"openai_compatible": {
"novita": {
"api_url": "https://api.novita.ai/openai/v1",
"available_models": [
{
"name": "moonshotai/kimi-k3",
"display_name": "Kimi K3 via Novita",
"max_tokens": 1048576,
"max_output_tokens": 131072,
"reasoning_effort": "high",
"capabilities": {
"tools": true,
"images": true,
"parallel_tool_calls": false,
"prompt_cache_key": false,
"chat_completions": true,
"interleaved_reasoning": false,
"max_tokens_parameter": true
}
}
]
}
}
}
}After saving, enter the key through Zed's provider interface. The 131072 output setting is a conservative editor cap below the model's published maximum; it does not reduce the 1M context setting.
VS Code: use Cline or Roo Code
Cline officially supports custom OpenAI-compatible providers, and Roo Code offers a similar provider-based workflow.
- Install Cline or Roo Code in VS Code.
- Open the extension's provider settings and choose OpenAI Compatible.
- Base URL: https://api.novita.ai/openai/v1.
- API key: use the key created in Novita.
- Model ID: moonshotai/kimi-k3.
- Save, verify the connection and begin with read-only file access.
- Test file edits and terminal tools separately before enabling automatic approvals.
The extension controls checkpoints, terminal access, MCP and approval rules. K3 is the model behind those features; the affiliate link does not install or configure the extension.
Neovim: use CodeCompanion
Novita publishes an official CodeCompanion.nvim integration for bringing its models into Neovim.
- Install CodeCompanion.nvim using its current official installation instructions.
- Create a Novita adapter in your Neovim configuration.
- Use https://api.novita.ai/openai as the adapter base URL unless the current plugin version asks for the /v1 form.
- Store your Novita key in an environment variable rather than directly in init.lua.
- Set the model to moonshotai/kimi-k3.
- Run :checkhealth codecompanion and test a small chat before editing code.
Neovim plugin configuration can change between releases. Compare this page with Novita's current CodeCompanion guide and the plugin's own adapter documentation.
Other IDEs and coding agents
A client does not need a dedicated Novita button if it supports a custom OpenAI-compatible or Anthropic-compatible provider.
- Look for Provider, Custom Model, OpenAI Compatible, Anthropic Compatible or Base URL in the client's settings.
- Use https://api.novita.ai/openai/v1 for an OpenAI-compatible connection or https://api.novita.ai/anthropic for an Anthropic-compatible connection.
- Enter moonshotai/kimi-k3 as the model ID and add your Novita API key.
- Confirm that the client uses a compatible chat or messages endpoint.
- Test chat first, then file edits, tools, terminal access, MCP and autocomplete one feature at a time.
Do not describe this as literally working in every IDE. A client must support a compatible custom endpoint, and each client decides which agent features it exposes.
Quick API test before opening a large repository
A small test confirms that the key, endpoint and model ID work. Run it only in a secure terminal and remove the environment variable when finished.
$env:NOVITA_API_KEY="PASTE_YOUR_KEY_HERE"
$headers = @{
Authorization = "Bearer $env:NOVITA_API_KEY"
"Content-Type" = "application/json"
}
$body = @{
model = "moonshotai/kimi-k3"
messages = @(
@{ role = "user"; content = "Return only: Kimi K3 connection works" }
)
max_tokens = 80
reasoning_effort = "low"
} | ConvertTo-Json -Depth 5
Invoke-RestMethod `
-Uri "https://api.novita.ai/openai/v1/chat/completions" `
-Method Post `
-Headers $headers `
-Body $bodyDo not save the real key in this command, screenshots, shell history, dotfiles or source control. Revoke and replace it immediately if it is exposed.
Before you start
- Confirm the checkout or top-up currency and current K3 token rates in your Novita account.
- Set a small initial budget and leave automatic top-up disabled until you know your usage.
- Remove secrets, customer data and private keys from any code or context sent to the model.
- Update your chosen coding client and confirm that it supports a custom OpenAI-compatible or Anthropic-compatible provider.
- Keep the API key out of settings.json, project files, screenshots and public repositories.
How to confirm it worked
- The coding client's model selector or configuration shows moonshotai/kimi-k3 through Novita.
- A one-file test returns a valid answer without a 401, 404 or model-not-found error.
- Novita's usage dashboard records the test request and a small token charge or voucher deduction.
- File, terminal, tool and MCP access are limited to what you deliberately authorize in the client.
- If you need autocomplete or tab suggestions, confirm whether the client uses K3 or a separate completion model for that feature.
Start with a few dollars
You do not have to commit to a plan to find out whether K3 suits your work. Run it for a week, read the usage dashboard, then decide.
Affiliate link - no exclusive discount attached.FAQ
More Software offers
I may receive a reward if you use a link or code below.

Kimi K3: $5 First Month
$5 first month for Kimi K3 plus 15 other coding models, then $10/month. The plan is capped; subscribing through the referral link adds a stated $5 credit.

Pay $11 Instead of $22 for Your First Month
Create lifelike voiceovers, dub videos, transcribe audio and build a Professional Voice Clone with 121,000 credits on ElevenLabs Creator.

Video Without a Camera
Type a script, pick from 125+ AI avatars and publish a narrated video in 80+ languages. The free plan covers 10 minutes a month.