Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

Qwen3 offers downloadable models with two reasoning modes

Alibaba Cloud’s Qwen team published the original Qwen3 family as downloadable models in several sizes. The April 2025 models can switch between a reasoning mode and a direct-answer mode. [2][3]

Model access
Downloadable in multiple sizes, including dense and mixture-of-experts models. [3]
License
Apache License 2.0 for the eight original release checkpoints. [1][3]

What the model does

On April 29, 2025, Qwen released six dense checkpoints and two mixture-of-experts checkpoints under Apache 2.0. The Qwen team later submitted a technical report describing the series. [3][4]

Why it matters

Developers can choose a checkpoint that fits their hardware and spend more time on reasoning only where their application benefits from it. [2][3]

Who can use it

Application developers can compare checkpoint sizes for writing, coding and multilingual tasks. Researchers can evaluate the published weights and read the technical report. [2][3][4]

Key dates

  1. Original Qwen3 weights released [3]
  2. Technical report first submitted [4]

Ways to use it

Use the same tasks to compare the direct and reasoning modes on an original Qwen3 checkpoint. Measure response time and review answers for the languages and code you work with. [2][3]

Model limitations

Qwen3 is a family, so memory needs, available context and responses depend on the exact checkpoint. The two-mode description applies to the original April release. [1][3]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source Qwen Team’s Qwen3 release history and checkpoint documentation
  2. Primary source Qwen/Qwen3-235B-A22B
  3. Primary source Qwen3: Think Deeper, Act Faster | Qwen
  4. Primary source Qwen3 Technical Report

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked