Alibaba's Qwen 3.5 27B Challenges GPT-5 on Performance and Price

Alibaba's Qwen 3.5 27B Challenges GPT-5 on Performance and Price

Alibaba has released Qwen 3.5 27B, a 27 billion parameter multimodal language model that posted competitive benchmark scores against OpenAI's GPT-5 while running locally at up to 90 tokens per second on consumer hardware. The model was released in late February 2026 and has quickly gained traction in the open-source AI community.

The model features a native context length of 262,144 tokens, extendable to 1 million, and supports text, image, and video inputs in a unified architecture. On key benchmarks, Qwen 3.5 27B achieves 86.1% on MMLU-Pro, 85.5% on GPQA Diamond, and 72.4% on SWE-bench Verified—matching or exceeding GPT-5 on software engineering tasks while significantly undercutting API costs.

Community testing has been enthusiastic. Users on Reddit's LocalLLaMA forum report running the model at 90 tokens per second on setups with consumer GPUs like the RTX 3090 Ti, outperforming GPT-5 on practical tasks like building functional applications from natural language prompts.

"Qwen 3.5 27B is the REAL DEAL," wrote one Reddit user who tested the model. "Beat GPT-5 on my first test." The model scored 74.9% on SWE-bench Verified compared to GPT-5's 72.4%, according to benchmark data from Rival.tips.

Open Weights, Apache 2.0 License

Unlike GPT-5, which requires API calls or expensive subscription plans, Qwen 3.5 27B is available with open weights under the Apache 2.0 license. This allows developers to run the model locally, fine-tune it for specific use cases, and deploy it without ongoing per-token costs.

The model also supports 201 languages and dialects, includes tool use capabilities, and offers a hybrid thinking mode for reasoning tasks. Technical features include Grouped-Query Attention, 64 layers, SwigLU activation, and optimizations for lower latency on mobile GPUs through sparse attention.

With its combination of strong benchmarks, local inference capability, and open-source licensing, Qwen 3.5 27B represents a significant challenge to the closed-model dominance of OpenAI and Anthropic in the mid-range LLM market.