Qwen is a family of models developed by the Qwen Team at Alibaba Group, not a single chatbot. Since its public releases began in 2023, the family has expanded from downloadable language-model weights into a sequence of model generations with different sizes and capabilities. Qwen3 added documented thinking and non-thinking modes and tool-call support; agent applications built around Qwen still need software to manage tools, memory, permissions and evaluation.
How Qwen developed
The chronology below reflects milestones recorded by the Qwen project, not a complete history of every model or internal release. The model names and dates are the team’s published record.
As an Amazon Associate I earn from qualifying purchases.
| Period | Recorded milestones | What changed |
|---|---|---|
| 2023 | Qwen-7B and Qwen-7B-Chat on August 3; an Int4 Qwen-7B-Chat release on August 21; Qwen-14B and Qwen-14B-Chat on September 25; finetuning support in September; subsequent Qwen1.x releases. | The public trail began with downloadable language models and chat variants, alongside quantized and finetuning-related releases. |
| 2024 | Qwen1.5 in February; Qwen1.5-MoE-A2.7B in March; Qwen2 in June; Qwen2.5 in September. | The family added a mixture-of-experts release, then moved through two major named generations. |
| 2025 | Qwen3 was announced in April. Qwen3-2507 refreshed releases followed in July and August. | Qwen3 documentation described thinking and non-thinking use, tool integration, and a range of model sizes. |
| 2026 | The Qwen3.8 repository records Qwen3.5 announcements beginning February 16, additional sizes in February and March, Qwen3.6 releases in April, and Qwen3.8 releases in August. | The published sequence continued through Qwen3.8. This chronology is current to the repository’s August 2026 announcements. |
This progression is best understood as a changing model family and ecosystem, rather than one chatbot steadily gaining features. A version name alone does not identify every model’s size, modality, license or deployment requirements; those details belong to the particular checkpoint and its documentation.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →What changed with Qwen3
Qwen3’s documented family includes dense and mixture-of-experts variants, spanning these listed sizes: 0.6B, 1.7B, 4B, 8B, 14B, 32B, 30B-A3B and 235B-A22B. The naming and available options give developers different model-scale choices, but the figures do not by themselves establish which checkpoint will be fastest, most capable or suitable for a particular workload.
#1 Best Overall
The Qwen Team describes Qwen3 as supporting thinking and non-thinking modes and integration with external tools in both modes. Its documentation also claims support for more than 100 languages and dialects. These are the team’s descriptions of the family; they should not be read as an independent ranking or a guarantee that every checkpoint performs equally across tasks or languages.
When does Qwen become an agent?
A model that can reason or produce a tool call is not, by itself, a complete operating agent. The application must interpret the model’s output, decide whether a requested action is allowed, invoke tools, return results to the model, and manage the interaction over time. Memory, permissions and evaluation also sit in the surrounding application, not in a model name alone.
Rank #2
Qwen-Agent is the Qwen Team’s framework for building agent applications around Qwen. Its project materials describe instruction following, tool use, planning and memory, and provide examples including a browser assistant, a code interpreter and custom assistants. The project updates document Qwen3 tool-call demonstrations, MCP cookbooks, Qwen3-Coder and Qwen3-VL tool-call demonstrations, and a Qwen3.5 agent example. That shows development of an agent ecosystem; it does not establish that every task is reliable or safe without application-level controls and testing.
For practical deployments, treat tool access as a permission boundary. An agent that can browse, run code or call external services may affect data and systems beyond the conversation. Select which tools it can use, constrain the actions those tools permit, and evaluate the complete application on representative tasks rather than inferring reliability from a model’s ability to emit tool-call syntax.
Open-weight does not mean one license for every Qwen model
The Qwen3 repository says, “All our open-weight models are licensed under Apache 2.0.” That statement applies to the open-weight models covered by that repository; it should not be generalized to every earlier Qwen checkpoint. The older Qwen repository records separate Tongyi Qianwen license agreements for early Qwen-72B, Qwen-14B and Qwen-7B checkpoints, with commercial use subject to the applicable agreement and application process. It describes different terms for Qwen-1.8B as well.
Before using a checkpoint commercially, inspect that specific model’s model card and attached license. A family name or a license statement for a newer generation is not a substitute for checking the terms that accompany the weights you intend to deploy.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Choosing a model and deployment route
The official materials establish multiple model sizes and local deployment pathways, but do not provide a buyer-grade head-to-head comparison of every current checkpoint. Choose against the intended workload and verify details for the named model version.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match- Scale and architecture: Compare dense and mixture-of-experts options against the memory, throughput and quality needs of your workload; parameter labels alone are not a performance comparison.
- Task and mode: Check whether the specific checkpoint supports the reasoning-oriented or faster general interaction pattern you need, and how its documented modes work.
- Modality: Confirm that the chosen release accepts the required inputs, such as text or images. Do not assume capabilities carry over unchanged between checkpoints.
- Context and serving: Match context length to actual prompt and output needs, then check how the selected runtime handles that workload.
- Hosted or local: Qwen documents multiple local inference and serving options, so local execution is possible but not required. A hosted route may simplify operations; a local route may better suit particular privacy or infrastructure needs. Weigh latency, data handling, cost and available hardware for your situation.
- Agent framework and governance: For tool-using applications, assess orchestration, tool permissions, memory, observability and evaluation alongside the model.
One Qwen Team Qwen3.8-27B serving command documents a context length of 262,144 tokens and tensor-parallel size four. Those figures describe that example configuration, not a promise of equal performance in every runtime or a universal minimum hardware requirement. Check the serving instructions for the exact checkpoint and system you plan to use.
Best Value
What the evolution means for developers
Qwen’s path runs from early downloadable language models through successive generations and, by the team’s published record, into tooling intended for agent applications. The significant shift is not simply that a model can answer questions or generate tool calls: developers now have a broader set of model and framework components to assemble. Whether a resulting application is capable, useful and safe depends on the particular checkpoint and the engineering around it.
The cited project documentation is primary evidence for what Qwen announced, supports and licenses. It does not establish that Qwen is superior overall to other model families; such a judgment requires a dated comparison of named checkpoints using appropriate tasks and evaluation methods.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
