Qwen3 8B↗
A compact text model for experiments with limited local GPU memory.
Compare non-thinking latency separately from reasoning. Keep reasoning tokens in the output budget.
Qwen
Open-weight text and vision-language models. Compare the exact checkpoint, thinking mode and serving configuration.
Official documentation ↗Qwen3 32B↗
A dense text model with thinking and non-thinking modes.
A dense deployment has different memory and serving requirements from an MoE model. Check the actual quantization and context configuration.
Qwen3 235B A22B↗
A mixture-of-experts text model in the Qwen3 family.
Active parameters do not equal the total memory needed to load all weights. Compare serving configurations explicitly.
Qwen2.5 VL 7B↗
A smaller vision-language model for image and document tasks.
Image resolution and preprocessing affect both resource use and output. Pin the image inputs and settings in an evaluation.
Keep publisher, model and endpoint separate
The publisher identifies the model family or checkpoint. An inference operator selects a runtime, quantization and service configuration. The API endpoint defines the accepted request shape and operational limits. Compare all three when planning a migration.
Track lifecycle changes
Official release notes can announce previews, general availability, retirement and alias changes. A model appearing in an API list is an observation of access, not necessarily a new public release.
Open the release radar →