Qwen3.8 27B
Text generationQwen3.8-27B
Qwen3.8 27B on FlexAI: Alibaba LLM, Apache 2.0 license, $0.2 / $2.5 per M tokens on the OpenAI-compatible Token Factory.
Pricing and availability
| Rate | Price |
|---|---|
| Input | $0.2 / M tokens |
| Output | $2.5 / M tokens |
| Cached input | $0.03 / M tokens |
| Availability | Serverless, on the shared Token Factory API |
Deployment requirements
- Context
- 256K tokens
- Capabilities
- Accepts image input (vendor-declared)
- API endpoint
- /v1/chat/completions
- Compatibility
- OpenAI
- Parameters
- 27.8B
- Active parameters
- 27.8B of 27.781428B
- License
- Apache 2.0
- Hardware
- 2× H100
- GPUs to self-host
- 2× H100
- Quantization (served)
- BF16
- Native precision
- BF16, from torch_dtype
- Weights
- Qwen/Qwen3.8-27B
- Published by
- Alibaba
- Origin
- China
- Agent-ready
- Tool calling (measured)
Estimate your monthly cost
Share of input tokens served from prompt cache: repeated system prompts, long documents, and multi-turn context.
Estimate only, at the current market-tracked rate. Usage-based; no minimums.
Get an API keyServerless or dedicated for Qwen3.8 27B
Below about 7.1B tokens a month, serverless is cheaper. Above it, a dedicated 2× H100 endpoint at $6,322/month costs less.
Assumes a 70/30 input/output token mix and an always-on endpoint at 730 hours a month. Dedicated bills per minute and autoscales, so serving less than continuously moves the crossover earlier. Your mix moves the crossover. Open the full calculator.
How to call Qwen3.8 27B
from openai import OpenAI
client = OpenAI(base_url="https://api.flex.ai/v1", api_key="your-api-key")
response = client.chat.completions.create(
model="Qwen3.8-27B",
messages=[{
"role": "user",
"content": [
{"type": "text", "text": "What's in this image?"},
{"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}},
],
}],
)
print(response.choices[0].message.content)Swap models, not your agent
Qwen3.8 27B runs behind the same OpenAI-compatible key as every other model in the catalog. Change the model string and your skills, tools, scoped memory, and governance carry over unchanged. The same setup scales to a dedicated endpoint or your own private cloud.