Kimi K3
Kimi K3 is Moonshot AI's newest open-weight flagship — a 2.8-trillion-parameter MoE with a 1M-token context and always-on deep reasoning, strong at agentic coding, terminal and tool use, and multimodal understanding. Access the Kimi K3 API on Velokey at a lower price — one API key, pay as you go, below official pricing.
Use cases
Long-horizon software engineering · Agents and business automation
Input
Text, image
Output
Text
Billing
Per token
Start a conversation
Type a message below to begin
Sign up to get $0.5 in free credits — about 20 free images
Pricing details
Transparent usage-based pricing with no hidden fees.
Detailed Pricing
Usage-based pricing breakdown
Kimi K3
| Spec | Price | Official/Fal Price(USD) | Savings |
|---|---|---|---|
| Input/1M tokens | 16 Credits≈$16 | 20 Credits | -20% |
| Output/1M tokens | 80 Credits≈$80 | 100 Credits | -20% |
| Cache Read/1M tokens | 1.6 Credits≈$1.6 | 2 Credits | -20% |
Billing Rules
- Output price varies by model, resolution, quality, duration, or token usage.
- Uploaded reference files may be charged separately when supported.
- Add-on capabilities such as web search or image search may be billed per request.
- Token-based models are billed by input and output tokens.
- If the primary route is unavailable, Velokey may automatically switch to a stable fallback route when possible.
- Usage is pay-as-you-go with transparent credit deduction.
* Final cost depends on the generated result.
Related models
Explore other models in the same family.
MoonshotSave 20%Kimi K2.5
Kimi K2.5 is Moonshot AI's visual agentic model with visual coding, four operating modes, and an Agent Swarm coordinating up to 100 sub-agents in parallel.
- MoonshotSave 20%
Kimi K2.6
Kimi K2.6 is Moonshot AI's model with state-of-the-art coding, Awwwards-level full-stack front-end generation, a 50+ agent swarm, and Document-to-Skills.
- MoonshotSave 20%
Kimi K2.7 Code
Kimi K2.7 Code is Moonshot AI's 1T-parameter (32B active) MoE coding model for long-horizon software engineering, with a 256K context and always-on thinking.
OpenAISave 20%GPT 5.4 Mini
OpenAI's fast, cost-efficient GPT-5.4 Mini for high-volume production, with vision, long context, and strong tool use at low latency.
Affordable Kimi K3 API
Access Moonshot AI's newest open flagship, Kimi K3, at a lower price on Velokey — 2.8T MoE, 1M-token context, agentic coding, and multimodality, OpenAI-compatible, one API key, pay as you go, below official pricing.
Meet Kimi K3 and its API integration
Kimi K3 is Moonshot AI's newest open-weight flagship — a 2.8-trillion-parameter MoE with a 1M-token context and always-on deep reasoning, strong at agentic coding, terminal and tool use, and multimodal understanding. On Velokey you can access the Kimi K3 API at a lower price — one API key, pay as you go, below official pricing, no forced subscription — with credits, usage, and multi-model routing all managed in one console.
2.8T open-weight MoE with always-on reasoning
Kimi K3 uses a 2.8-trillion-parameter MoE (896 experts, 16 activated per token) and runs always-on chain-of-thought at maximum reasoning effort by default, giving grounded answers on complex reasoning, math, and knowledge work; weights are released under a Modified MIT license, so you can self-host or call it directly on Velokey.
View docs →
1M-token long context
A 1M-token context window holds large codebases, long documents, and multimodal material in a single session, with flat pricing across the full window and no long-context surcharge — a strong fit for retrieval, synthesis, and long-horizon tasks over lengthy material.
View docs →
Agentic coding and agent swarm
Kimi K3 treats terminal use and tool calling as first-class, is compatible with OpenAI function/tool-call schemas, and supports plan mode, todo lists, skill invocation, nested agents, and parallel sub-agent swarms — taking on multi-hour engineering sessions and long-horizon software tasks, and ranking among the top on frontend-code and SWE coding benchmarks.
View docs →
Native multimodal understanding
Native support for text, images, and video, plus direct parsing of PDF, Word, and Excel documents — a fit for bringing screenshots, reports, and mixed-content material into a single pass of reasoning, with fewer separate parsing and conversion steps.
View docs →
How to deploy the Kimi K3 API on Velokey
A few steps to start calling.
Register and get an API key
Create an API key in the Velokey console; one key authenticates every request sent to Kimi K3.
Set the model
Set model to kimi-k3 and submit messages and tools (function calling) as needed.
Call and integrate
Send requests through Velokey's unified /v1/chat/completions endpoint, compatible with the OpenAI calling convention, with usage and credits managed in one console.
Kimi K3 real-world use cases
From long-horizon engineering to research and analysis, a fit for many high-autonomy scenarios.
Long-horizon software engineering
With terminal use, codebase navigation, and agent swarms, take on multi-hour, multi-step engineering tasks from spec to working code.
Agents and automation
Use OpenAI-compatible tool calling to wire Kimi K3 into tool chains and build multi-step agents that plan autonomously and run in parallel.
Research and analysis
Always-on reasoning plus a 1M-token context suits retrieval, synthesis, and deep analysis over long documents, charts, and code.
Multimodal document processing
Native understanding of images and video, with direct parsing of PDF, Word, and Excel, bringing screenshots, reports, and mixed content into a single pass.
Data analysis and insight
Reason in multiple steps over structured and unstructured data to produce conclusions, summaries, and decision recommendations.
Knowledge work and content
Long-document understanding, structured output, and content generation as a daily assistant for engineering and research teams.
Why choose Velokey for Kimi K3
Below official pricing
Call Kimi K3 at a lower price on Velokey — pay as you go, no forced subscription, lower upfront cost.
One key, many models
A single API key reaches Kimi K3 and other vendors' models, turning task routing and cross-model fallback into configuration rather than rewrites.
Unified OpenAI-compatible access
Compatible with the OpenAI calling convention; switching models is a single model-field change, keeping integration cost minimal.
Stability and high availability
Automatic failover and stable fallback routes maintain availability when the primary route is down, keeping production steady.
API Reference
Complete API call examples and parameter descriptions
Endpoint
https://api.velokey.ai/v1/chat/completionsAuthentication
Bearer YOUR_API_KEYRequest Example
curl https://api.velokey.ai/v1/chat/completions \
-X POST \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"messages": [{"role": "user", "content": "Hello!"}],
"temperature": 0.7
}'Response Example
{
"id": "req_abc123",
"model": "kimi-k3",
"created": 1234567890,
"data": { ... }
}Kimi K3 API FAQ
What is the Kimi K3 API?
Kimi K3 is Moonshot AI's newest open-weight flagship — a 2.8-trillion-parameter MoE with a 1M-token context and always-on deep reasoning, strong at agentic coding, tool use, and multimodal understanding. On Velokey you call it through unified OpenAI-compatible access, with model set to kimi-k3.
What makes Kimi K3 stand out?
It's billed as the largest open-weight model (2.8T MoE, 896 experts with 16 activated), with always-on deep reasoning, first-class terminal and tool use, parallel sub-agent swarms, and native understanding of text, images, video, and PDF/Word/Excel — ranking among the top on frontend-code and SWE coding benchmarks.
How long is Kimi K3's context?
1M tokens (about 1,048,576), enough to hold large codebases, long documents, and multimodal material in a single session, with flat pricing across the full window and no long-context surcharge.
Does Kimi K3 support function calling and agents?
Yes. Kimi K3 makes tool calling and terminal use first-class, compatible with OpenAI function/tool-call schemas, and supports plan mode, skill invocation, and parallel sub-agent swarms — a fit for long-horizon agentic workflows that plan autonomously.
How is it billed on Velokey?
The pricing section on this page shows current Velokey prices, billed by input and output tokens (about $16 input / $80 output per 1M tokens, official $20/$100). Credits are managed from one Velokey account.
Start building with Kimi K3 today
Use one API key to access low-cost, stable AI models through Velokey.