Model facet · API

Qwen3.8-2.4T-A95B API: 3 Reasoning Levels, OpenAI-Compatible

Qwen3.8-2.4T-A95B exposes OpenAI- and Anthropic-compatible endpoints so most existing SDKs and coding tools can point at it with a base URL change. Thinking mode is mandatory on this checkpoint — set reasoning_effort explicitly to control cost.

Two request protocols

ProtocolUse whenBase path
OpenAI Chat CompletionsMost existing OpenAI-SDK integrations and coding toolsView
Anthropic MessagesClaude Code and other Anthropic-SDK-based toolsView

Minimal request

Example
curl https://api.oneinfer.ai/v1/ula/chat/completions \
  -H "Authorization: Bearer $ONEINFER_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-2.4t-a95b","messages":[{"role":"user","content":"Reply with: connection verified"}],"reasoning_effort":"low"}'

Thinking mode is mandatory

The Qwen3.8-2.4T-A95B model card states thinking mode is required and cannot be disabled on this checkpoint. Every response opens with a think block. Use reasoning_effort: low to reduce cost on simple requests, or switch to Qwen3.8-Max if you need a non-thinking-mode variant of the same weights.

Three reasoning levels

LevelUse whenCost note
lowSimple Q&A, classification, short completionsCheapest tier; lowest think-block spend
mediumDefault for most agentic workloadsBalanced cost and reasoning depth
xhighHardest reasoning, multi-step planning, SWE workHeaviest think-block spend; pair with batch or cache where possible

Ready to test the workflow?

Create account & add credits

Live model identifier pending OneInfer registry

The exact model identifier routed through OneInfer will be confirmed once Qwen3.8-2.4T-A95B is added to the OneInfer registry. Use the slug above as the model parameter; the rest of the request shape (messages, reasoning_effort, sampling parameters) follows the OpenAI Chat Completions API.

Go deeper

  • Full setup and troubleshooting for a specific key or tool — see the integration guides.
  • Defining tools and validating structured output — see the function-calling guide.
  • Choosing an effort level per task — see the reasoning-effort guide.

Frequently asked questions

Can I disable thinking on Qwen3.8-2.4T-A95B?

No. The model card states thinking mode is required and cannot be disabled on this checkpoint. Use reasoning_effort: low to reduce cost on simple requests, or switch to Qwen3.8-Max, which supports non-thinking mode.

What reasoning_effort levels are available?

Three levels: low, medium, and xhigh. The default at most providers is medium. Lower levels reduce the size of the think block and therefore reduce the per-request token cost.

Where is the live Qwen3.8-2.4T-A95B model page?

The canonical model page with current OneInfer pricing, capabilities, and availability is /models/Qwen/Qwen3.8-2.4T-A95B. This page is a focused facet of that entity, not a replacement for it.

How should I treat benchmark or price claims?

Check each claim's provenance label and observed date. Vendor-reported and independently verified numbers are shown as separate evidence classes on this hub.

Is Qwen3.8-2.4T-A95B open source?

Yes — weights are released under the qwen3.8-max license. Confirm the license tag against the official Hugging Face repository before self-hosting at scale.

Put Qwen3.8-2.4T-A95B to work

Fund a controlled evaluation, start with a prepared prompt, and measure quality and cost on your own workload.