2.8T parameters MoE · #1 on MMLU-Pro at 89.2% · 256K context window
Any developer can now download weights or call via API
| Metric | K3 | GPT-5 | DeepSeek V4 |
|---|---|---|---|
| MMLU-Pro | 89.2% | 87.8% | 88.1% |
| HumanEval+ | 94.7% | 93.2% | 92.8% |
| GSM8K | 96.4% | 95.1% | 95.6% |
| Context Window | 256K | 128K | 128K |
| API Input Price | $0.50/M | $2.50/M | $0.50/M |
| API Output Price | $2.00/M | $10.00/M | $2.00/M |
Source: Artificial Analysis, July 2026
K3 uses 2.8T total / 32B active MoE architecture. Recommended setup:
Weights on Hugging Face: moonshotai/kimi-k3
No GPUs, no config, no waiting for downloads. Call K3 (and 5 other top Chinese models) through a single TokenEase API endpoint.
curl -X POST https://tokenease.io/api/register \
-H "Content-Type: application/json" \
-d '{"email":"your@email.com"}'
Get $1 worth of tokens (1M tokens) valid for 14 days. No credit card required.
from openai import OpenAI
client = OpenAI(
api_key="your-tokenease-key",
base_url="https://tokenease.io/v1"
)
response = client.chat.completions.create(
model="kimi-k3",
messages=[{"role": "user", "content": "Explain quantum computing like I'm 5"}]
)
print(response.choices[0].message.content)
models = ["kimi-k3", "deepseek-chat", "glm-4-plus", "qwen-plus"]
for model in models:
response = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": "Write a Python function to reverse a string"}]
)
print(f"{model}: {response.choices[0].message.content[:50]}...")
| Feature | Moonshot Direct | OpenRouter | TokenEase |
|---|---|---|---|
| K3 Input Price | $0.50/M | $0.75/M | $0.50/M |
| K3 Output Price | $2.00/M | $3.00/M | $2.00/M |
| Single Endpoint | No | Yes | Yes |
| Overseas Developers | CN phone required | Supported | No CN phone needed |
| Multi-Model | Separate accounts | Supported | One key, all models |
| Auto-Failover | No | Limited | 6-provider backup |
| Free Trial | Limited | No | 1M tokens |
1M tokens · 14 days · No credit card