Data policyUpdated November 18, 2026

Silk Compute data retention, explained

Silk Compute does not store your prompts or model outputs.

2 source pages
Prompts kept by default
Zero data retention
RetentionReadNov. 18, 2026
Training on API prompts
Prompts and responses are never stored.
TrainingReadNov. 18, 2026
Keeping nothing takes
A verified email

No approval step or contract

Zero retentionReadNov. 18, 2026
Rank
#1of 10

See the ranking

RankingReadNov. 18, 2026

On the timeline

Default retention of API prompts and outputs
Not storedKept by defaultKept longer if flagged
Silk ComputeOpen-weight models
Zero data retention
abliteration.aiOpen-weight models
Never stored by default
FeatherlessOpen-weight models
Not logged
Unfil.AIOpen-weight models
Up to 24 hours
xAI APIGrok models
30 days
OpenAI APIGPT models
Up to 30 days
Claude APIClaude models
Within 30 daysUp to 2 years if flagged2 years if flagged
MorphOpen-weight models
Up to 30 days
Gemini APIGemini models
55 days
Mistral APIMistral models
No period published
Log scale, read November 18, 2026.
PRIVATE AI APICC-BYSources for every bar

What the policy says

Silk ComputeZero data retention
"Silk Compute does not store your prompts or model outputs."

It names no plan, tier or approval step.

silkcompute.com
Silk ComputeData handling
"Prompts and responses are never stored."
silkcompute.com
Silk ComputeModel page
"Prompts and model outputs pass through our servers and are not stored. We keep only billing records, such as token counts and cost."

Every model page shows this line next to the price.

silkcompute.com

Strengths and shortfalls

Strengths

  • All six models read 1,048,576 tokens.
  • It accepts the OpenAI and Anthropic API formats.

Shortfalls

  • It serves open-weight models only, not GPT, Claude or Gemini.
  • No moderation layer, so your app sets its own content rules.

Models

All six support tool calling and streaming. Prices are per 1M tokens, read November 18, 2026.

ModelModel idInputOutput
Kimi K3.1 ProMoonshot AI, 2,780B parameterskimi-k3.1-pro-uncensored$3.00$8.00
DeepSeek V4.1 Flash 1017DeepSeek, 763B parametersdeepseek-v4.1-flash-1017-uncensored$0.24$0.90
DeepSeek V4.1 Flash 1017 FastDeepSeek, 763B parametersdeepseek-v4.1-flash-1017-fast-uncensored$0.24$0.90
GLM-5.3Z.ai, 753B parametersglm-5.3-uncensored$2.00$4.00
GLM-5.3 FlashZ.ai, 320B total, 18B active parametersglm-5.3-flash-uncensored$0.24$1.00
Hy4Tencent, 770B (preview card) parametershy4-uncensored$1.60$3.20

Kimi K3.1 Pro is the strongest for security research and red teaming, and DeepSeek V4.1 Flash 1017 has the lowest blended price.

Endpoints

The base URL https://api.silkcompute.com answers POST /chat/completions, POST /v1/messages, POST /responses, so the OpenAI and Anthropic SDKs work with a new base URL and key.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.silkcompute.com",
    api_key="YOUR_SILK_COMPUTE_KEY",
)
reply = client.chat.completions.create(
    model="deepseek-v4.1-flash-1017-uncensored",
    messages=[{"role": "user", "content": "Summarize this contract."}],
)
print(reply.choices[0].message.content)

For Claude Code, set these variables, as the Claude Code guide explains.

export ANTHROPIC_BASE_URL=https://api.silkcompute.com
export ANTHROPIC_AUTH_TOKEN=YOUR_SILK_COMPUTE_KEY
export ANTHROPIC_MODEL=kimi-k3.1-pro-uncensored
claude

Plans and payment

Plans include a monthly usage allowance, and prepaid credits from $5 never expire. You can pay by card or with Monero, Bitcoin, Ether, USDT or USDC.

PlanMonthlyYearlyUsage included each month
Go$10$100$12
Pro$20$200$25
Max$50$500$70
Ultra$200$2000$300

Full prices are on the Silk Compute pricing page.

Other API policies