Operational difference
Direct
Fast, economical responses for assistants, RAG, classification, extraction and interactive flows.
Thinking
More internal tokens and more latency for planning, complex analysis or multi-step tasks.
Latency cost
In internal benchmarks, thinking was roughly 10x slower and generated many more tokens than direct. Tessera therefore keeps it outside the TTFT SLA. There is no quota: it is unlimited on every tier.
| Mode | Recommended use | TTFT SLA | Quota |
|---|---|---|---|
| Direct | Interactive production | Covered by tier | Included |
| Thinking | Occasional reasoning | Not covered | Unlimited |
How to pick the mode
Change the model field. Direct: Qwen/Qwen3.6-35B-A3B or Qwen/Qwen3.8-27B. Thinking: Qwen/Qwen3.6-35B-A3B-thinking or Qwen/Qwen3.8-27B-thinking. The gateway sets the reasoning mode explicitly on each alias, so you don't need extra_body and it works the same with any OpenAI-compatible client. Don't use chat_template_kwargs.enable_thinking or reasoning_effort to switch modes: the supported contract is the alias.
client.chat.completions.create(
model="Qwen/Qwen3.6-35B-A3B-thinking",
messages=[{"role": "user", "content": "Evaluate this plan"}],
)