Model · Moonshot AI
Kimi K3.1 Pro for reverse engineering
Kimi K3.1 Pro is the best model for reverse engineering. This abliterated build holds 1,048,576 tokens, writes up to 943,718, and reads text, images and video.
On this page
Specs
| Kimi K3.1 Pro | |
|---|---|
| Model id | kimi-k3.1-pro-uncensored |
| Maker | Moonshot AI |
| Parameters | 2,780B |
| Context | 1,048,576 tokens |
| Max output | 943,718 tokens |
| Reads | Text, images, video |
| Tool calling | Yes |
| Reasoning effort | low, high, max |
| Refusal behavior | Removed (abliterated) |
| Price per 1M tokens | $3.00 in, $0.30 cached, $8.00 out (Silk Compute, read November 18, 2026) |
Moonshot AI describes the original as "an open-weight, natively multimodal agentic model" with "a 1-million-token context window" on its Hugging Face model card. The comparison has the other four.
Calling it
Silk Compute serves it on Chat Completions, Anthropic Messages and Responses.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.silkcompute.com",
api_key=os.environ["SILK_API_KEY"],
)
response = client.chat.completions.create(
model="kimi-k3.1-pro-uncensored",
messages=[{"role": "user", "content": "Here is a decompiled function. Explain what it does."}],
)
print(response.choices[0].message.content)A plugin uses https://api.silkcompute.com, or https://api.silkcompute.com/v1 when it asks for the /v1 path, and kimi-k3.1-pro-uncensored. Model page on Silk Compute.
In an agent
This file sets Kimi K3.1 Pro as the default and lists the other models under /model.
{
"apiKeyHelper": "echo $SILK_API_KEY",
"env": {
"ANTHROPIC_BASE_URL": "https://api.silkcompute.com",
"ANTHROPIC_DEFAULT_MODEL": "kimi-k3.1-pro-uncensored",
"ANTHROPIC_DEFAULT_FABLE_MODEL": "kimi-k3.1-pro-uncensored",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "kimi-k3.1-pro-uncensored",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "kimi-k3.1-pro-uncensored",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4.1-flash-1017-uncensored",
"CLAUDE_CODE_MAX_CONTEXT_TOKENS": "1048576"
},
"modelPicker": {
"replaceBuiltInOptions": true,
"options": [
{ "model": "kimi-k3.1-pro-uncensored", "label": "Kimi K3.1 Pro" },
{
"model": "deepseek-v4.1-flash-1017-uncensored",
"label": "DeepSeek V4.1 Flash 1017"
},
{
"model": "deepseek-v4.1-flash-1017-fast-uncensored",
"label": "DeepSeek V4.1 Flash 1017 Fast"
},
{ "model": "glm-5.3-uncensored", "label": "GLM-5.3" },
{ "model": "glm-5.3-flash-uncensored", "label": "GLM-5.3 Flash" },
{ "model": "hy4-uncensored", "label": "Hy4" }
]
}
}Then add GhidraMCP, ReVa, IDA MCP, Binary Ninja MCP or JADX AI MCP. Codex and OpenCode are on the MCP setup page.
Prompts
Work from the leaf functions. For each, give a name, a one-line summary and a confidence.
Cost
200,000 tokens in and 50,000 tokens out cost about $1.00 at Silk Compute, 0.2 × $3.00 + 0.05 × $8.00, read November 18, 2026. A cached follow-up is $0.30 per 1M tokens. For thousands of small functions, DeepSeek V4.1 Flash 1017 is $0.24 in and $0.90 out.
Limits
Check every name against the code. The malware analysis guide covers sample text.