Model family

Qwen model family

Every Qwen model on OneInfer, from the 3.8 Flash and Max siblings back through 3.7, 3.6, and 3.5 checkpoints, plus the Qwen3 base, Coder, Next, and VL lines. This page tracks the family so an existing Qwen integration knows exactly what changes before switching checkpoints.

Qwen 3.8 line

VersionStatusPositioning
Qwen3.8-FlashCurrent Flash tierLow-latency Flash sibling at $0.16 / $0.47 per 1M input/output tokens, multimodal image + text input, 1M context.
Qwen3.8-MaxCurrentSame 1M context window, 12.5× higher input price, stronger reasoning on fourteen of fifteen benchmarks where Claude Opus 4.6 publishes a score.

Qwen 3.7 line

VersionStatusPositioning
Qwen3.7-MaxCurrentLarger Qwen 3.7 checkpoint; verify exact context and pricing per provider.

Qwen 3.6 line

VersionStatusPositioning
Qwen3.6-35B-A3BCurrentMoE checkpoint with 35B total / 3B active parameters; verify pricing per provider.
Qwen3.6-27BCurrentDense 27B checkpoint; lighter alternative to the 3.7 / 3.8 lines.

Qwen 3.5 line

VersionStatusPositioning
Qwen3.5-397B-A17BCurrentLargest MoE in the 3.5 line (397B total / 17B active).
Qwen3.5-122B-A10BCurrentMoE checkpoint at 122B total / 10B active.
Qwen3.5-35B-A3BCurrentMoE checkpoint at 35B total / 3B active.
Qwen3.5-27BCurrentDense 27B checkpoint; lighter alternative to the larger 3.5 MoEs.

Qwen 3 base line

VersionStatusPositioning
Qwen3-235B-A22B-Instruct-2507CurrentMoE instruction checkpoint at 235B total / 22B active.
Qwen3-235B-A22B-Thinking-2507CurrentMoE reasoning checkpoint at 235B total / 22B active.
Qwen3-30B-A3B-FP8CurrentSmaller MoE at 30B total / 3B active, FP8 weights.
Qwen3-8bCurrentDense 8B checkpoint for lightweight workloads.

Qwen3-Coder line

VersionStatusPositioning
Qwen3-Coder-480B-A35B-InstructCurrentLargest coding-specialised MoE at 480B total / 35B active.
Qwen3-Coder-NextCurrentNext-iteration coding checkpoint; smaller and faster.

Ready to test the workflow?

Create account & add credits

Qwen3-Next line

VersionStatusPositioning
Qwen3-Next-80B-A3B-InstructCurrentNext-line MoE instruction checkpoint at 80B total / 3B active.
Qwen3-Next-80B-A3B-ThinkingCurrentNext-line MoE reasoning checkpoint at 80B total / 3B active.

Qwen3-VL line

VersionStatusPositioning
Qwen3-VL-235B-A22B-ThinkingCurrentVision-language MoE at 235B total / 22B active.

Qwen image

VersionStatusPositioning
Qwen-image-txt2imgCurrentText-to-image generation model.

Flash vs Flash-Next: which is on this page?

Qwen3.8-Flash on this page is the hosted multimodal flash-tier API (Alibaba, 26 Aug 2026, image + text input, $0.16 / $0.47 per 1M tokens). Qwen3.8-Flash-Next is a separate release — the open-weights architecture preview for Qwen4, with 125B total parameters (6B active) under the qwen-community-1.0 license. The two should not be confused.

Deciding whether to switch checkpoints

Reproduce your existing Qwen workload against Qwen3.8-Flash on a fixed prompt set before switching production traffic — even within the same family, smaller Flash variants can shift verbosity and tool-use behavior.

Frequently asked questions

How is Qwen3.8-Flash different from Qwen3.8-Flash-Next?

Qwen3.8-Flash is the hosted multimodal flash-tier API model. Qwen3.8-Flash-Next is a separate open-weights architecture preview for Qwen4, released under qwen-community-1.0 — different model, different license, different deployment story.

How is Qwen3.8-Flash different from the full Qwen 3 line?

Qwen3.8-Flash is the smaller, faster sibling in the Qwen 3 family, optimized for low-latency tool use. The full Qwen 3 checkpoints offer larger context and stronger reasoning at higher cost.

Where is the live Qwen3.8-Flash model page?

The canonical model page with current OneInfer pricing, capabilities, and availability is /models/Qwen/Qwen3.8-Flash. 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.

Put Qwen3.8-Flash to work

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