Skip to content
For authorized security work

Open-weight models that answer the security question you actually asked.

Batchwell gives red teams, pentesters, malware analysts and detection engineers one API key for uncensored open-weight models. You pay per token from prepaid credit and see the cost of every request.

Accounts are for people 18 and over doing lawful work they are authorized to do. See the acceptable use policy.

curl https://api.batchwell.ai/v1/chat/completions \
  -H "Authorization: Bearer $BATCHWELL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "batchwell/deepseek-v4.1-flash-uncensored",
    "messages": [
      {
        "role": "system",
        "content": "Authorized internal pentest, scope ACME-2026-14."
      },
      {
        "role": "user",
        "content": "Which log sources would catch this Kerberoasting run?"
      }
    ],
    "models": ["batchwell/qwen3.6-72b-uncensored"]
  }'

Built around the work, not around refusals

General-purpose assistants are tuned to decline anything that resembles an attack. The models we host are open-weight checkpoints with those refusals removed, so they stay useful on authorized engagements.

Red team and pentest

Draft attack plans, payload notes and reports without a model refusing because the topic sounds offensive.

Malware analysis

Walk through decompiled functions, unpack loaders and explain obfuscation in plain language.

Exploit research

Reason about crash triage, root causes and patch diffs for vulnerabilities you are cleared to study.

Detection engineering

Turn techniques into Sigma, YARA and SIEM queries, then ask how an adversary would slip past them.

One endpoint, predictable behaviour

OpenAI-compatible

Point the OpenAI SDK you already use at our base URL. Chat completions, streaming and tool calls work as you expect.

Fallbacks before output

List backup models and provider preferences. If an upstream fails before the first token, we move on; once text starts, we never switch silently.

Cost on every response

Each response carries its exact charge in micro-USD and a generation id you can look up later. Prepaid credit, no subscription.

A narrow, published guard

Security content is allowed. We block only sexual content, anything involving minors, weapons and violence incitement, impersonation and CBRN uplift.

What we keep, and what we don't

We do not store your prompts or completions. We keep request metadata (model, token counts, cost, timing, status) so you can audit spend, and we keep content our guard flags so a person can review it.

Read the privacy notice

Prepaid credit, refundable for 7 days

Top up from $20 to $2,000 at a time. Credit never expires, and unused credit can be refunded within 7 days of purchase. The price you see per model already includes our fees.

How billing works