Qwen3.5 35B-A3B: Three Billion Active, Apache 2.0

What Qwen3.5 35B-A3B is

Qwen3.5 35B-A3B is the smallest mixture-of-experts model in Alibaba's Qwen3.5 generation: 35 billion total parameters with 3 billion active per token, Apache 2.0 and self-hostable, with a 256K token context window. At roughly $0.25 per million input tokens and $2 output through third-party providers it sits in the middle of the generation's pricing. Qwen3-Next-80B-A3B has the same three billion active parameters.

Other A3B models

Alibaba publish several models activating three billion parameters per token: this one has 35 billion total, Qwen3-Next-80B-A3B has 80 billion and Qwen3-30B-A3B-Instruct-2507 has 30 billion. On LMArena's text leaderboard Qwen3.5 35B A3b has 1394 Elo (rank 173); by category, 1435 Elo in Coding (rank 179), 1398 Elo in Math (rank 168), 1419 Elo in Expert (rank 161) and 1343 Elo in Creative Writing (rank 192). Qwen3-Next-80B-A3B-Instruct has 1399 Elo (rank 167) and Qwen3-30B-A3B-Instruct-2507 has 1382 Elo (rank 191).

What it inherits from the generation

Like the rest of Qwen3.5 it is built on a unified vision-language foundation using early fusion training on multimodal tokens, which Alibaba say outperforms the separate Qwen3-VL models across reasoning, coding, agents and visual understanding. It pairs Gated Delta Networks with sparse mixture-of-experts — the mechanism behind the low activation ratio — and covers 201 languages and dialects. Reinforcement learning was scaled across million-agent environments with progressively harder task distributions.

Alibaba’s figures

Alibaba's mid-size comparison table covers GPT-5-mini, GPT-OSS-120B, Qwen3-235B-A22B and the three Qwen3.5 mid-size models. Reading the rows the card labels: LongBench v2 gives Qwen3.5-122B-A10B 60.2, Qwen3.5-27B 60.6 and Qwen3.5-35B-A3B 59.0, against GPT-5-mini's 56.8 and GPT-OSS-120B's 48.2; HMMT February 2025 gives 91.4, 92.0 and 89.0 against 89.2 and 90.0; HMMT November 2025 gives 90.3, 89.8 and 89.2 against 84.2 and 90.0; and LiveCodeBench v6 gives 78.9, 80.7 and 74.6 against 80.5 and 82.7. Qwen3-235B-A22B records 54.8, 85.1, 89.5 and 75.1 on the same four. These are Alibaba's own figures.

Architecture and hosted version

The card lists 40 layers with a hidden dimension of 2,048, arranged as 10 repeats of three Gated DeltaNet layers followed by one gated-attention layer, each feeding a mixture-of-experts block of 256 experts with 8 routed and 1 shared active and an expert intermediate dimension of 512. The gated-attention layers use 16 query heads and 2 key-value heads. Alibaba state that Qwen3.5-Flash, on Alibaba Cloud Model Studio, is the hosted version of these weights, with a 1M context by default and official built-in tools. The repository holds the post-trained model in Hugging Face Transformers format, compatible with Transformers, vLLM, SGLang and KTransformers. The context is 262,144 tokens natively, extensible to 1,010,000. The token embedding is padded to 248,320 entries. Alibaba also describe near-100% multimodal training efficiency compared with text-only training, and asynchronous reinforcement-learning frameworks for large agent scaffolds.

The Qwen3.5 family

The Qwen3.5 open-weight releases covered on this site are the 397B-A17B flagship, the 122B-A10B and 35B-A3B mixture-of-experts models, and the dense 27B. Alibaba name hosted versions for two of them on Alibaba Cloud Model Studio: Qwen3.5-Plus for the 397B-A17B and Qwen3.5-Flash for the 35B-A3B. The models can also be tried in Qwen Chat at chat.qwen.ai.

Official Resources