# Partition Labs

> Flat-rate LLM inference. A subscription buys a guaranteed share of GPU capacity,
> plus prepaid pay-as-you-go credits, behind an OpenAI- and Anthropic-compatible API.

Service tiers set queue admission order, never stream speed: every tier runs the same
model at the same tokens per second, and a higher tier starts sooner when the pool is
busy. Prefix-cache writes are free.

## Base URLs

Keep the SDK you already use and change one environment variable:

```
OPENAI_BASE_URL=https://api.partitionlabs.ai/v1
ANTHROPIC_BASE_URL=https://api.partitionlabs.ai
```

Streaming, tool calls and image input work on both surfaces. Batches and files use the
OpenAI shapes (`/v1/files`, `/v1/batches`).

## Models

- `qwen3.8-27b-fp8-4k` — Qwen 3.8 27B FP8, 4k context pool (4k context)
- `qwen3.8-27b-fp8-8k` — Qwen 3.8 27B FP8, 8k context pool (8k context)
- `qwen3.8-27b-fp8-16k` — Qwen 3.8 27B FP8, 16k context pool (16k context)
- `qwen3.8-27b-fp8` — Qwen 3.8 27B FP8, General pool (32k context)
- `qwen3.8-27b-fp8-64k` — Qwen 3.8 27B FP8, 64k context pool (64k context)
- `qwen3.8-27b-fp8-128k` — Qwen 3.8 27B FP8, 128k context pool (128k context)
- `qwen3.8-27b-fp8-256k` — Qwen 3.8 27B FP8, 256k context pool (256k context)

## service_tier

Pass `service_tier` on a request (OpenAI: top-level; Anthropic: same field). The tier
that served the request is echoed back in `usage`.

- `priority` — Priority: queue-jump under contention
- `standard` — Standard: the default
- `flex` — Flex: same model and speed; last in line
- `batch` — Batch: async; completes within 24h or 7d

## Pages

- https://partitionlabs.ai/pricing — subscription ladder per pool and the pay-as-you-go rate card
- https://partitionlabs.ai/status — measured time-to-first-token per band, last 24 hours