Qwen/Qwen3-4B-Instruct-2507

Qwen3-4B-Instruct-2507 is a compact multilingual text instruction model with a reported 262k-token context window. Its Artificial Analysis Intelligence Index of 7 and 4B scale support a materially lower technical-capability assessment than Gemini 3.5 Flash, Kimi K2.5, GPT-5.5, and Claude Opus 4.8. It is text-oriented in the supplied material; there is no verified image, audio, video, or structured tool-I/O evidence, so multimodal scoring remains near the catalog floor rather than comparable multimodal anchors. Coding evidence is limited but concrete: the cited H4 report gives the base model a 35.1 LiveCodeBench result; its 40.3 figure applies to a separately Codeforces-SFT-trained variant and is not credited here. That places it above the minimally evidenced Pegasus coding profile but well below Kimi K2.5, GPT-5.6 Luna, and frontier coding anchors. No exact-model DeepSWE score is listed, and supplied SWE-bench, Terminal-Bench/Aider, and LMArena/Arena-Hard checks do not provide a usable result. Reasoning is therefore conservative. The official Qwen documentation, Hugging Face presence, and public GitHub repository make local integration and inspection comparatively straightforward. Artificial Analysis lists $0.00 per million input/output tokens, giving unusually strong potential cost effectiveness, but official provider pricing and an output-speed measurement were not supplied; this sharply limits pricing and availability confidence. Public release visibility is credible but weaker evidence of adoption than the major proprietary anchors.