Qwen 3.8 27B Flex: the same model on spare capacity, at a fraction of the price

Qwen 3.8 27B Flex is now available as a public model on QDivZero. It is the same Qwen 3.8 27B we already serve: a dense 27B model with a 256k-token context window, image input, tool calling and an OpenAI-compatible API. The difference is how we price it.

Flex runs on capacity that Unlimited Basic already has available but is not using at that moment. Instead of leaving that capacity idle, we offer it at €0.08 per million input tokens and €0.29 per million output tokens. The trade-off is simple: when the capacity is busy, a Flex request may take longer to start. There is no latency guarantee.

The same model, not a smaller one

Flex is not a distilled or cut-down version of Qwen 3.8 27B. It is the same model and the same request contract. You can use it with the model identifier public:qwen/qwen-3.8-27b-flex through the same OpenAI-compatible chat completions endpoint as the rest of the public catalog.

The 256k context window makes it useful for long documents, repository-sized prompts and multi-step agent traces. The model keeps Qwen's reasoning controls (low, medium, xhigh) and the multimodal input supported by the endpoint. Nothing about the model changes because the price is lower.

Why Flex can cost less

Unlimited Basic includes unlimited Qwen 3.8 27B, subject to fair use. Serving that plan requires capacity for the moments when subscribers need it. Most of the time, that capacity is not fully occupied.

Flex turns that spare capacity into a second service level. It is not a temporary promotion, and it is not a different model. It is a way to use capacity that would otherwise sit idle, with a price that reflects the lower priority.

The practical consequence matters more than the mechanism:

• When capacity is free, Flex requests run normally.
• When capacity is busy, Flex requests can queue or take longer to start.
• There is no latency SLA. If your application needs a predictable response time, use Unlimited Basic or the standard public endpoint.

Flex is paid per token from your account balance. It does not require an Unlimited subscription, and it does not consume AI Credits.

Price comparison

The comparison uses the public per-token prices at the time of publication. A request with 1 million input tokens and 1 million output tokens costs €0.37 with Flex versus €4.01 with the standard public endpoint. For a workload that generates 10 million input tokens and 2 million output tokens, the difference is €1.38 versus €16.12.

Check the public pricing page for the current rates. Prices can change, and your actual cost depends on the input and output mix of each request.

How Flex compares with other Qwen 3.8 27B providers

To put the rate in context, we checked the Qwen 3.8 27B endpoints listed on OpenRouter on 2 November 2026 and converted the EUR prices at EUR/USD 1.116:

Flex is not the cheapest option for input tokens. Several providers list lower input rates. It does have the lowest output rate we found in that comparison. Because output tokens usually cost more and often dominate long generations, the cheapest total depends on the input and output mix of your workload.

OpenRouter aggregates many providers, but not the entire market. Prices and exchange rates change, so treat this as a dated reference rather than a permanent ranking.

When Flex is the right fit

Flex is designed for work where a slower response is acceptable in exchange for a much lower token price:

• Batch generation. Drafts, summaries, translations, classifications and content variants that run in the background.
• Evaluations and synthetic data. Large prompt sets where total cost matters more than the latency of any single request.
• Long-running agents. Tasks that already wait on tools, retrieval or human review, where a queue delay is a small part of the total time.
• Cost-sensitive prototypes. Trying Qwen 3.8 27B on real data without committing to a subscription.
• Off-peak workloads. Jobs that can run when demand is lower and capacity is more likely to be free.

When Unlimited Basic is the better fit

Choose Unlimited Basic when the model is part of an interactive product. The subscription includes unlimited Qwen 3.8 27B subject to fair use, with limits of 60 requests per minute per account or API key and 5 concurrent requests to the model. It also includes 5,000 shared monthly AI Credits for GLM-5.3-Flash, DeepSeek V4.1 Flash and DeepSeek V4 Flash 0731.

That plan is the better choice for coding assistants, chat interfaces and agents that need to respond quickly. Flex is the better choice when the workload can wait and the bill is the constraint.

How to use Flex

Send a normal chat completions request with the Flex model identifier:

1curl https://api.qdiv0.com/v1/chat/completions \
2  -H "Authorization: Bearer $QDIV0_API_KEY" \
3  -H "Content-Type: application/json" \
4  -d '{
5    "model": "public:qwen/qwen-3.8-27b-flex",
6    "messages": [
7      {
8        "role": "user",
9        "content": "Summarise this changelog in five bullets."
10      }
11    ],
12    "stream": true
13  }'
14

The request shape is the same one you already use for other public models. If your client supports OpenAI-compatible endpoints, change the model identifier and keep the rest of the integration.

What to expect

Flex is a best-effort service. Capacity comes from the Unlimited pool, so performance depends on what subscribers are doing at that moment. A request that starts immediately in a quiet period may queue during a busy one. The price is lower because we are not reserving capacity for it.

That makes Flex a poor fit for hard real-time requirements, interactive chat with tight latency targets, or anything that needs a contractual response time. It is a good fit for the work that can wait five seconds, five minutes or five hours without changing the outcome.

Try Qwen 3.8 27B Flex

You can try the model in the QDivZero chat demo and check the current token rates on the public pricing page. If you already have an account, use the model identifier public:qwen/qwen-3.8-27b-flex with your API key. If you do not, create an account and start with a small workload.

The same model, on capacity that would otherwise be idle, at a fraction of the price.