Can you own it?
This page is a projection of the one entry record, the Use & modify and Transparency factors that Assess covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.
- AssessUse & modify + Transparency
- ImplementData control + Reliability
- UseReliability
- SupportTransparency
Intended & out-of-scope use
Qwen3 is Alibaba's general-purpose open-weight family, strongest in coding and agentic work. Qwen3-Coder-480B-A35B is among the strongest open coding models. Intended use spans assistant, coding and tool-using applications across sizes from 0.6B laptop models to a 480B coder.
Parts of the family are out of scope. Built-in alignment is lighter than at Western frontier labs, so deploy behind your own guardrails. The systemic-risk large variants (235B, 480B-Coder) carry an unclosed EU AI Act Article 55 gap, so they are out of scope for EU high-stakes use unless the deployer closes that gap.
Known limitations, bias & failure modes
Qwen3 shows China-aligned topic censorship: the models decline or steer on subjects sensitive under Chinese content rules, which matters for any knowledge or retrieval use.
Instruct alignment withstands casual jailbreaks but is lighter than at Western frontier labs, and no companion guard or classifier model exists. Capability leadership is concentrated in coding and agentic tasks and is not uniform across domains.
Openness tier & components
Qwen is open_weights: Apache-2.0 weights and a good model card, with closed training data and training code and only partially reproducible evaluation. The model can be run and adapted, but its training cannot be reproduced.
License terms & permitted use
The mainstream Qwen3 sizes are OSI-approved Apache-2.0 with unconditional commercial use, which is more permissive than the restrictive community licences some peers use. Licensing is not uniform across the family: some large or legacy variants use the Tongyi Qianwen community licence. Verify the LICENSE file per checkpoint.
Supply-chain & provenance
Qwen is distributed from the verified Qwen org on Hugging Face and officially dual-published on Alibaba's ModelScope hub. Checkpoints are safetensors with checksums, and no canonical-org malicious incident is on record.
Dual-hub distribution is a supply-chain surface in itself, so verify that checksums match across Hugging Face and ModelScope. There is no cryptographic signing or provenance attestation.
EU AI Act posture
Qwen is a GPAI model. The mainstream Apache-2.0 sizes plausibly qualify for the open-source exemption. The 235B MoE and Qwen3-Coder-480B likely exceed the 1e25 FLOPs systemic-risk threshold, and for systemic-risk models the exemption is void and the full Article 55 package is owed.
That documentation is not published, and a China-based provider is unlikely to furnish an EU AI Office package. An EU deployer of the large variants therefore has a compliance gap that upstream artifacts cannot close.
Benchmarks & evaluation
Third-party evaluation shows strong general capability and class-leading open coding (Qwen3-Coder). The results are third-party and the evaluation harness is only partially open, so treat the numbers as indicative and not as verified results.
How this scores
The ownership factors this domain covers, drawn from the one entry record.
Use and modify freelyCan you run, modify and adapt it with no gate and no field-of-use trap?
StrongThe licence is per checkpoint. Mainstream Qwen3 (0.6B, 32B, 235B-A22B, Coder-480B-A35B-Instruct) is Apache-2.0 with unconditional commercial use and no MAU clause. The legacy and large Tongyi Qianwen checkpoints (Qwen-72B, Qwen2.5-72B-Instruct) carry a >100M-monthly-active-user licence trigger, a 'Built with Qwen' attribution requirement and export controls. Check the LICENSE file on the exact checkpoint before relying on Apache terms.
TransparencyDo you know what it is: weights, training, behaviour, and legible terms?
WeakThe weights are held but cannot be inspected: training data and code are closed and China-aligned topic censorship is built into behaviour, so sensitive-topic coverage is distorted in ways that cannot be audited. On the hosted path, the binding Alibaba Cloud Product Terms (Sec 4.48, updated 19 Oct 2026) say Alibaba Cloud 'will not use your Member Content to develop or improve the models on Model Studio, unless you separately provide your consent'. The FAQ repeats a softer 'never uses your data for model training'. The Terms state no retention period, storage region or opt-out mechanism.
Sources
The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.