Qwen3.5 27B by Alibaba: dense open-weight model with early fusion vision-language, Gated DeltaNet hybrid architecture, 262K context, and 201-language support.
Capabilities, design details, and architectural traits
Qwen3.5 27B is the dense open-weight variant in the Qwen3.5 series. Unlike its MoE siblings (35B-A3B, 122B-A10B, 397B-A17B), every parameter is active during inference. It includes a built-in vision encoder, making it a natively multimodal causal model rather than a language-only backbone with a separately attached vision module.
| Trait | Detail |
|---|---|
| Unified vision-language via early fusion | Text, image, and video tokens are interleaved and trained jointly from pretraining, not via post-hoc adapters - the official model card explicitly states this achieves cross-generational parity with Qwen3 and outperforms Qwen3-VL on vision benchmarks |
| Multi-Token Prediction (MTP) | Trained with multi-step MTP, documented in the model card as a distinct training objective |
| Asynchronous RL at million-agent scale | Reinforcement learning training used asynchronous frameworks supporting massive-scale agent scaffolds across progressively complex task distributions |
| 201 languages | Documented multilingual coverage across 201 languages and dialects - notably broader than the 119 documented for Qwen3 |
| Context window | 262,144 tokens natively, extensible to 1,010,000 via YaRN |
| License | Apache 2.0 open-weight |
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Alibaba | Anthropic | Anthropic |
| Release Date | February 24, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.30 Best Input Pricing | $5 | $10 |
| Output Pricing | $2.40 Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | textimagevideo | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 30.0 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | - | 78.0 Best Coding Index | 76.5 |
| Agentic Index | - | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
