Kimi K3 API
Moonshot's 2026 flagship, released July 16. A 2.8-trillion-parameter open-weight MoE with a 1M context window and native vision — the largest open-weight model in the world at release.
💰 Save up to 70% vs official Moonshot AI pricingKimi K3 — cheaper than the official Moonshot AI API
Access Kimi K3 through AI API Hub and pay less per token. Same OpenAI-compatible endpoint, lower cost.
Technical Specifications
| Provider | Moonshot AI |
| Model Family | Kimi K3 |
| Release Date | 2026-07 |
| Context Window | 1M |
| Max Output Tokens | 1,000,000 |
| Input Price | $3.00 / 1M tokens |
| Output Price | $15.00 / 1M tokens |
| Vision Support | Yes ✓ |
| Function Calling | No |
| JSON Mode | No |
| Streaming | Yes ✓ |
| Fine Tuning | Not Available |
| Status | Active ✓ |
Overview
Kimi K3 is Moonshot AI's current kimi model, released in 2026-07. Moonshot's 2026 flagship, released July 16. A 2.8-trillion-parameter open-weight MoE with a 1M context window and native vision — the largest open-weight model in the world at release.
For developers, the headline numbers are a 1M context window and up to 1,000,000 output tokens per response — enough headroom for 2.8t open weights and 1m context without chunking your input. Priced at $3.00/1M input and $15.00/1M output, it sits in the premium tier — best reserved for tasks where quality justifies the spend.
On the capability side, Kimi K3 exposes 4 features: Vision, Long Context, Reasoning, Streaming. Note that fine-tuning isn't supported — you'll work with the base model. Vision support means you can pass images alongside text, handy for document parsing or UI automation.
The practical appeal of routing Kimi K3 through AI API Hub is simplicity: one OpenAI-compatible endpoint, USDT & USDC payments, no credit card, and you're calling the API in under 30 seconds — just swap your base URL.
What Makes Kimi K3 Different
How Kimi K3 is used
Kimi K3 is deployed for problems that need structured thinking — math derivations, logic puzzles, scientific analysis, and multi-step research synthesis. The reasoning chain adds latency, so reserve it for tasks where a standard chat model produces shallow or incorrect answers. For routine Q&A and summarization, a cheaper sibling model is more cost-effective.
Pricing position within Moonshot AI
Kimi K3 is the most expensive active model from Moonshot AI at $3.00/1M input — the premium tier. The cheapest sibling costs $0.56/1M (81% less). Reserve Kimi K3 for tasks where its capabilities justify the premium; route bulk traffic to cheaper siblings.
Kimi K3's role in the lineup
Within Moonshot AI's lineup, Kimi K3 is the flagship — the most capable and most expensive option. It's part of the kimi family (3 active variants share this family). Use it when quality matters more than cost; switch to a cheaper sibling for high-volume or less demanding tasks.
Real-world use cases
Real-world deployments: math problem solving, scientific research synthesis, multi-step logical analysis, and structured decision support systems. Kimi K3 excels where standard chat models produce shallow answers — the reasoning chain catches edge cases that faster models miss.
vs sibling models
What makes Kimi K3 different from sibling models: compared to Kimi K2.6 ($2.10/1M cheaper, 128K vs 1M context (smaller)); Kimi K2.5 ($2.44/1M cheaper, 128K vs 1M context (smaller)). Choose Kimi K3 when step-by-step reasoning matters more than speed.
API Examples
Python
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.apiyihe.org/v1"
)
response = client.chat.completions.create(
model="kimi-k3",
messages=[
{"role": "user", "content": "Hello"}
]
)
print(response.choices[0].message.content)JavaScript / Node.js
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.API_KEY,
baseURL: "https://api.apiyihe.org/v1"
});
const response = await client.chat.completions.create({
model: "kimi-k3",
messages: [
{ role: "user", content: "Hello" }
]
});
console.log(response.choices[0].message.content);cURL
curl https://api.apiyihe.org/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "kimi-k3",
"messages": [
{"role": "user", "content": "Hello"}
]
}'Supported Features
| Vision / Image Input | ✅ Supported |
| Audio / Voice Input | ❌ Not Available |
| Function Calling | ❌ Not Available |
| JSON Mode | ❌ Not Available |
| Streaming | ✅ Supported |
| Fine-Tuning | ❌ Not Available |
| Multimodal | ❌ Not Available |
Benchmark Scores
| Benchmark | Score |
|---|---|
| MMLU | Not Publicly Available |
| GPQA | Not Publicly Available |
| SWE-Bench | Not Publicly Available |
| HumanEval | Not Publicly Available |
| GSM8K | Not Publicly Available |
| MATH | Not Publicly Available |
| MMMU | Not Publicly Available |
Pricing History
Kimi K3 was released in 2026-07 by Moonshot AI and is currently publicly available via AI API Hub.
Current Pricing: $3.00 per 1M input tokens · $15.00 per 1M output tokens. Pay-as-you-go with no minimum commitment.
Pricing Model: Token-based billing (pay per use). No subscription fees. No hidden costs.
💡 Moonshot AI occasionally updates pricing. AI API Hub reflects current pricing in real-time. All prices in USD. Pay with USDT or USDC — no currency conversion fees.
Compare Alternatives
Frequently Asked Questions
What is Kimi K3?
Kimi K3 is Moonshot AI's current kimi model. Moonshot's 2026 flagship, released July 16. A 2.8-trillion-parameter open-weight MoE with a 1M context window and native vision — the largest open-weight model in the world at release. It offers a 1M context window and supports Vision, Long Context, Reasoning. You can access it through AI API Hub using USDT or USDC — no credit card required.
How much does Kimi K3 cost?
Kimi K3 is priced at $3.00 per 1M input tokens and $15.00 per 1M output tokens, billed pay-as-you-go with no minimum. Through AI API Hub you can start with as little as $5 and scale from there.
Kimi K3 vs GPT-5.6 Sol?
They're built for different jobs. Kimi K3 costs $3.00/1M input with a 1M window; GPT-5.6 Sol runs $5.00/1M input with 1M. Kimi K3 is the more cost-effective pick and still brings 2.8t open weights. See the full side-by-side at /compare/kimi-k3-vs-gpt-5.6-sol/.
Kimi K3 context window?
Kimi K3 has a 1M context window, capable of processing up to 1,048,576 tokens in a single request. Maximum output tokens: 1,000,000.
Does Kimi K3 support function calling?
No, Kimi K3 does not natively support function calling. For function calling use cases, consider Moonshot AI's flagship models.
Is Kimi K3 multimodal?
Partially — Kimi K3 supports vision (image input) but not native audio processing.
Kimi K3 API rate limits?
Kimi K3 rate limits: 2K RPM. Higher tier plans offer increased throughput. For high-volume production use, consider Moonshot AI's faster variant models.
How to access Kimi K3 API?
Access Kimi K3 through AI API Hub: (1) Register at api.apiyihe.org/register?aff=8JZC, (2) Deposit USDT/USDC, (3) Get your API key instantly, (4) Use the OpenAI-compatible endpoint https://api.apiyihe.org/v1 with model name "kimi-k3". Start building in under 30 seconds.