Qwen3.8-27B API pricing and specifications
US-hosted, OpenAI-compatible Qwen3.8-27B inference with a tested 262,144-token context, transparent per-token pricing, and measured capacity.
Model card
| Public model ID | qwen/qwen3.8-27b |
|---|---|
| API | OpenAI-compatible chat completions — POST {base}/chat/completions (base = https://prefillsystems.com/v1) |
| Usage reporting | Prompt, cached-prompt, and completion counts reported by the inference worker |
| Authentication | Bearer API key for completions; model metadata is public |
| Access | Controlled direct API beta · request access · OpenRouter listing pending approval |
| Quantization & context | NVFP4 weights · FP8 KV cache · 262,144-token context · Qwen3 tokenizer |
| Capabilities | Image understanding · tool calls · structured outputs · reasoning; streaming and non-streaming |
Full model metadata is public at GET {base}/models; chat completions require a bearer key.
Pricing
Rates are per one million tokens (USD) and match the live gateway accounting configuration exactly:
| Token type | USD per 1M tokens | Notes |
|---|---|---|
| Input — uncached prompt | $0.24 | Applied to prompt tokens not reported as cached by the engine |
| Input — cached prompt | $0.12 | Applied to engine-reported cached prompt tokens when the backend reports them |
| Output — completion | $2.15 | Applied to generated completion tokens |
List models
Read the public model document without authentication:
curl https://prefillsystems.com/v1/models
Response (trimmed for readability; the live document also carries description, supported_parameters, and per-fleet capacity entries):
{
"data": [
{
"schema_version": "2.4",
"id": "qwen/qwen3.8-27b",
"name": "Prefill Systems: Qwen3.8 27B",
"hugging_face_id": "Qwen/Qwen3.8-27B",
"created": 1788566400,
"quantization": "nvfp4",
"tokenizer": "Qwen3",
"input_modalities": [
{
"type": "text",
"supported_inputs": {
"max_context_length": { "value": 262144, "unit": "token" }
},
"pricing": [
{ "type": "prompt", "unit": "token", "cost_usd": "0.00000024" },
{ "type": "cached_prompt", "unit": "token", "cost_usd": "0.00000012", "implicit": true }
],
"capacity": [
{ "type": "prompt", "unit": "token", "per": "minute", "value": 85000 }
]
},
{
"type": "image"
}
],
"output_modalities": [
{
"type": "text",
"max_length": { "value": 262144, "unit": "token" },
"streaming": true,
"pricing": [
{ "type": "completion", "unit": "token", "cost_usd": "0.00000215" }
],
"capacity": [
{ "type": "completion", "unit": "token", "per": "minute", "value": 6500 }
]
}
],
"is_ready": true,
"openrouter": { "slug": "qwen/qwen3.8-27b" },
"datacenters": [ { "country_code": "US" } ],
"deployment_region": "US",
"compliance": { "zdr": true },
"capacity": [
{ "type": "request", "unit": "request", "per": "minute", "value": 10 },
{ "type": "concurrency", "unit": "request", "value": 4 }
]
}
]
}Pricing is per token (cost_usd); the table above is the per-million equivalent.
For client setup and function calling, use the OpenAI Python SDK guide and Qwen3.8 tool-calling guide. Performance and conformance evidence is published in the endpoint benchmark report.
Honest limits
- We are a modest provider of GPU inference capacity, with plans to scale. Current capacity — 85,000 prompt tokens and 6,500 completion tokens per minute, 10 requests per minute, 4 concurrent requests, set below what we measured — is declared on the model document at GET https://prefillsystems.com/v1/models and changes as the fleet changes.
- No uptime commitment. The pilot has no SLA. Live state is on the status page.