Specs & substance
- Developed by
- Moonshot
- Model family
- Kimi
- Use case
- Multimodal agentic
- Modality
- Text + Vision
- Context window
- 1M tokens
- Architecture
- MoE + KDA + Gated DSA
- Version
- K3.1 Pro
- License
- Kimi K3.1 Pro License
- Pricing
- $2.00 in · $10.00 out · $0.20 cache read
- Released
- Nov 2026
- Endpoint
- parasail-kimi-k31-pro
Kimi K3.1 Pro is an open-weight, natively multimodal agentic model and Moonshot’s most capable 3T-class model to date. Post-trained from Kimi K3 with large-scale joint reinforcement learning across hundreds of downstream domains, it is the first model in the K3.1 series, delivering substantial improvements in coding, 3D creation, financial research, factual accuracy, and other agentic capabilities over a 1M-token context window.
The architecture keeps K3’s 2.8T total parameters with 104B activated, 93 layers of 69 KDA attention layers, and the MoonViT-V2 vision encoder — but replaces the 24 full-attention Gated MLA layers with Gated DeepSeek Sparse Attention (Gated DSA) layers for more efficient long-context modeling. Weights are MXFP4 with MXFP8 activations from quantization-aware training, and thinking effort is configurable per request with a `reasoning_effort` of low, high, or max.
Parasail serves Kimi K3.1 Pro on a global fleet of current-gen GPUs behind a single OpenAI-compatible endpoint — with per-token pricing, no minimums, and dedicated capacity options when you need guaranteed throughput.
Measured on the work that matters
Kimi K3.1 Pro model-card results alongside selected comparison models.
All scores are self-reported by Moonshot in the Kimi K3.1 Pro model card on Hugging Face ; the comparison columns are those Moonshot reports in its model card; coding and agentic evaluations use Claude Code / Kimi Code harnesses, with each benchmark’s provenance noted there. Where a benchmark is reported both with and without tool augmentation, we show the lower, no-tools number.
Drop-in via the OpenAI SDK
Point any OpenAI-compatible client at Parasail and change the model name. That's it.
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.parasail.io/v1"
)
response = client.chat.completions.create(
model="parasail-kimi-k31-pro",
messages=[
{"role": "user", "content": "Implement Hello World in Python"}
],
stream=True,
stream_options={"include_usage": True},
max_tokens=1000,
extra_body={"reasoning_effort": "max"}
)
for chunk in response:
if chunk.choices and chunk.choices[0].delta.content is not None:
print(chunk.choices[0].delta.content, end="", flush=True)from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY",
base_url="https://api.parasail.io/v1"
)
response = client.chat.completions.create(
model="parasail-kimi-k31-pro",
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": "What is broken in this UI screenshot?"},
{"type": "image_url", "image_url": {"url": "https://example.com/screenshot.png"}}
]
}
],
max_tokens=1000
)
print(response.choices[0].message.content)curl https://api.parasail.io/v1/chat/completions \
-H "Authorization: Bearer $PARASAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "parasail-kimi-k31-pro",
"messages": [{"role": "user", "content": "Implement Hello World in Python"}],
"stream": true,
"max_tokens": 1000
}'Explore the library
Kimi K3
Moonshot's 2.8T-parameter open-weight multimodal agentic model — frontier reasoning, native vision, and a 1M-token context window.
Kimi K2.7 Code
Specialized coding model optimized for software engineering and agentic coding workflows.
GLM-5.3
Z.AI’s strongest GLM-5 agentic coder — same base as 5.2, all the gains from post-training, open-source SOTA on coding and cyber benchmarks.
GLM-5.3 Flash
Z.AI’s first natively multimodal GLM-5 model — 320B total / 18B active parameters at a tenth of GLM-5.2’s price.
Run Kimi K3.1 Pro on Parasail.
Call it on an OpenAI-compatible endpoint with per-token pricing, or ask us about dedicated capacity.