Open-Weight Models Compared: Llama 4, Qwen3, DeepSeek, Mistral, Gemma 4, and gpt-oss

Open-weight model selection is not a single leaderboard. Compare quality on your workload, license, context behavior, tool calling, hardware cost, and maintenance burden.nnA reproducible shortlist should use the same prompts, quantization, context limits, and acceptance tests. The “best” model is the one that meets the product constraint with a clear fallback path.

Sources

See the primary sources linked in the Chinese edition.