Qwen3.5
A model series released by Alibaba's Tongyi Lab under the Qwen family, running in parallel with Qwen3. Detailed parameter specifications have not been fully disclosed, and the series may have connections to the Qwen2.5 line of models.
Core Facts
Qwen (the Tongyi Qwen series) is a family of open-source large models launched by Alibaba.
Qwen3.5's parameter activation rate is 4.3%, while DeepSeek V3's 671B model has an activation rate of 5.5%.
Qwen3.5 uses a hybrid attention architecture that mixes Gated Delta Net (a linear attention variant) with Gated Attention (standard attention).
Timeline (last 90 days)
On M4 Max, Omni 3B has a median dialogue generation time of 171 ms and a peak process RSS of 2.57 GiB; Qwen3.5 0.8B has a median dialogue generation time of 96 ms and a peak process RSS of 1.08 GiB; Qwen3-TTS 1.7B has a synthesis time of 1.78 seconds and a peak process RSS of 3.85 GiB.
The small model pair halves the download size and reduces peak memory by about 35% compared with the original model pair, but in this trial the full turn was actually slower.
Qwen3.5, Qwen3.6, and Qwen3.8 gained faster prefill and speculative decoding support on MLX.
In evaluations across 3 seeds, 3 models, and 4 benchmarks (involving Gemma 4, Qwen3.5, and Magistral), U-Lens is the best estimator, ahead of 5-round sampling, 10-round sampling, and supervised baselines.
NVIDIA Nemotron 3、Qwen3.5 至 Qwen3.8、Kimi-K3 以及 GLM-5.3-Flash 等模型均采用了混合架构范式。
Qwen3.5 was released under the Apache 2.0 license, permitting commercial use and post-training modifications.
The Qwen3.5 flagship 397B-A17B adopts a hybrid architecture that combines Gated Delta Networks with sparse Mixture-of-Experts (MoE), activating only 17 billion parameters per token.
Qwen3.5 scored 86.7 on TAU2-Bench and 85.0 on MMMU.
The research team fine-tuned the Qwen3.5 series of small language models on filtered and rebalanced teacher-generated supervised data.
Both the Clef model and the Laya model with a Qwen3.5 backbone can be exported to GGUF format and served via llama.cpp, including multimodal decision models capable of understanding images.
21 more timeline events
All Facts (20)
Qwen (the Tongyi Qwen series) is a family of open-source large models launched by Alibaba.
65%VerifiedQwen3.5's parameter activation rate is 4.3%, while DeepSeek V3's 671B model has an activation rate of 5.5%.
65%VerifiedQwen3.5 uses a hybrid attention architecture that mixes Gated Delta Net (a linear attention variant) with Gated Attention (standard attention).
65%UnverifiedOn M4 Max, Omni 3B has a median dialogue generation time of 171 ms and a peak process RSS of 2.57 GiB; Qwen3.5 0.8B has a median dialogue generation time of 96 ms and a peak process RSS of 1.08 GiB; Qwen3-TTS 1.7B has a synthesis time of 1.78 seconds and a peak process RSS of 3.85 GiB.
50%UnverifiedThe small model pair halves the download size and reduces peak memory by about 35% compared with the original model pair, but in this trial the full turn was actually slower.
50%UnverifiedQwen3.5, Qwen3.6, and Qwen3.8 gained faster prefill and speculative decoding support on MLX.
50%UnverifiedIn evaluations across 3 seeds, 3 models, and 4 benchmarks (involving Gemma 4, Qwen3.5, and Magistral), U-Lens is the best estimator, ahead of 5-round sampling, 10-round sampling, and supervised baselines.
50%UnverifiedNVIDIA Nemotron 3、Qwen3.5 至 Qwen3.8、Kimi-K3 以及 GLM-5.3-Flash 等模型均采用了混合架构范式。
50%UnverifiedQwen3.5 was released under the Apache 2.0 license, permitting commercial use and post-training modifications.
50%UnverifiedQwen3.5 scored 86.7 on TAU2-Bench and 85.0 on MMMU.
50%UnverifiedThe Qwen3.5 flagship 397B-A17B adopts a hybrid architecture that combines Gated Delta Networks with sparse Mixture-of-Experts (MoE), activating only 17 billion parameters per token.
50%UnverifiedThe research team fine-tuned the Qwen3.5 series of small language models on filtered and rebalanced teacher-generated supervised data.
50%UnverifiedBoth the Clef model and the Laya model with a Qwen3.5 backbone can be exported to GGUF format and served via llama.cpp, including multimodal decision models capable of understanding images.
50%UnverifiedThis version supports exporting Clef models based on the Qwen3.5 backbone, as well as Laya models, to GGUF format.
50%UnverifiedAfter integrating with Pi Coding Agent, simple repository queries are handled by Qwen3.5 while complex code generation requests are routed to Gemma, with the token rate stable in the 80–95 range.
50%UnverifiedIn the demo, when querying a simple factual question the router selects the Qwen3.5 4-billion-parameter small model, while complex tasks are handled by the Gemma model at a speed of approximately 93.1 tokens per second.
50%UnverifiedQwen3.5 with approximately 800 million parameters achieved the fastest generation speed, approaching 15 tokens per second, with the second-place LFM2.5 exceeding 10 tokens per second.
50%UnverifiedIn the summarization task, 4 out of the 8 models failed: the 4-billion-parameter Qwen3.5 fabricated content, LFM2.5 altered facts, Bonsai added nonexistent information, and MiniCPM5 missed key points.
50%UnverifiedIn Qwen3.5, GPT-OSS, and OLMo-3.1, removing the "source endorsement" direction using the mean difference method reduced compliance with incorrect sources by 64 to 78 percentage points, while removing the "user endorsement" direction reduced compliance by at most 11 percentage points.
50%UnverifiedThe Jeff project distilled and fine-tuned two base models, Qwen3.5 and Gemma 4, into decision models with 0.8B and 2B parameters respectively.
50%