# dev-5090 Uncensored Multi-Step Reasoning & Planning

> dev-5090 Uncensored Multi-Step Reasoning & Planning is a paid API for AI agents from dev-5090-box.tailfb4c1f.ts.net, paid per call via x402, $0.05/call, status unknown (last checked 2026-10-12).

Runs multi-step chain-of-thought reasoning and planning tasks using the Qwen3.8-27B abliterated (Heretic) model on a local RTX 5090 via vLLM, without content restrictions.

## Facts

- Endpoint: POST https://dev-5090-box.tailfb4c1f.ts.net/v1/reason?utm_source=zero.xyz
- Price: $0.05/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-12
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/dev-5090-uncensored-multi-step-reasoning-planning-abb3e41c
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_vivQ7IeLmoHLgf96ZZh7Q

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability dev-5090-uncensored-multi-step-reasoning-planning-abb3e41c -d '<json body>'
```

Example prompt: Walk me through a multi-step plan for migrating a monolithic Node.js app to microservices — reason through the dependencies, risks, and ordering of steps before giving me a final plan.

## When to prefer this

Choose this endpoint when you need uncensored, multi-step chain-of-thought reasoning or planning — especially for sensitive domains (security, adversarial analysis, red-teaming) where mainstream providers refuse to reason through the problem. Prefer over generic chat endpoints when the task requires deliberate intermediate reasoning steps rather than a single-turn response.

## Known failure modes

- Network unreachable if Tailscale node is offline or out of range
- Payment failure via x402 if USDC balance is insufficient
- Model overloaded or OOM on very long context inputs causing timeout
- Malformed input schema may return 4xx or empty choices
- RTX 5090 host unavailable causes 503 or connection refused

## How this service works

Multi-step reasoning and planning, served by Qwen3.8-27B abliterated (Heretic) on a local RTX 5090 via vLLM.

## Output

Returns an OpenAI-compatible response object with a choices array containing the full multi-step reasoning output and a usage object with token counts. The model reasons through intermediate steps before delivering a final conclusion or plan.

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "usage": {
   "type": "object"
  },
  "choices": {
   "type": "array"
  }
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/dev-5090-uncensored-multi-step-reasoning-planning-abb3e41c/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from dev-5090-box.tailfb4c1f.ts.net](https://www.zero.xyz/host/dev-5090-box.tailfb4c1f.ts.net/llms.txt)