[Qwen 3 8B Instruct](/models/alibaba--qwen-3-8b-instruct) and [Yi 1.5 9B Chat](/models/01-ai--yi-1.5-9b-chat) are similar in size but differ in training generation — Qwen 3 is the newer architecture with updated instruction tuning, while Yi 1.5 is a well-established model with a large community footprint. Both price in the $0.03–0.07/M token range, making cost roughly equivalent across providers. The differentiation is quality and language coverage. Qwen 3 8B Instruct scores approximately 72–74% on MMLU versus Yi 1.5 9B Chat's 65–68%, a 5–8 point gap that reflects the training generation difference. Qwen 3 8B also has tighter instruction-following — structured output tasks like JSON extraction and step-by-step reasoning with explicit formatting constraints have a higher success rate out of the box, reducing the need for few-shot examples. Yi 1.5 9B Chat has strong Mandarin Chinese performance relative to its parameter count, making it a practical choice for Chinese-language applications where 01.AI's training priorities pay off. The model also has a wider selection of community fine-tunes for domain-specific use cases (legal, medical, customer support), which matters if you're building on a fine-tuning workflow. Pick Yi 1.5 9B Chat if Chinese-language quality is the primary requirement or if you're building on an existing Yi-based fine-tune ecosystem. Pick Qwen 3 8B Instruct for English-dominant workloads where MMLU ceiling, instruction-following reliability, or structured output fidelity drives model selection.