Qwen3 Coder Next
Qwen3 Coder Next by Alibaba: open-weight MoE coding agent with 3B active of 80B total params, hybrid attention architecture, 256K context, and FIM support.
Model Overview
Capabilities, design details, and architectural traits
Qwen3 Coder Next - Hybrid Attention MoE Built for Local Coding Agents
Qwen3 Coder Next is an open-weight MoE model built on top of Qwen3-Next-80B-A3B-Base, a base that introduces a hybrid attention architecture combining Gated DeltaNet (linear attention) layers and Gated Attention layers interleaved with Mixture-of-Experts blocks. It runs only in non-thinking mode - no IMD blocks are generated, making it a direct-output model designed for low-latency agent loops.
| Trait | Detail |
|---|---|
| Architecture | Hybrid layout: 12 cycles of 3x (Gated DeltaNet -> MoE) followed by 1x (Gated Attention -> MoE) |
| Active vs. total parameters | 3B activated out of 80B total, with 512 experts and 10 activated per token |
| Non-thinking mode only | Does not generate IMD blocks; enable_thinking=False is no longer required |
| Context length | 262,144 tokens natively |
| Fill-in-the-Middle (FIM) | Supported for code insertion tasks across all Qwen3-Coder variants |
| Recovery from execution failures | Documented training objective; the model is agentic-trained at scale on executable code tasks |
| License | Apache 2.0 open-weight |
Scaffold-Agnostic CLI/IDE Integration
Qwen3 Coder Next is explicitly designed to adapt to multiple scaffold templates, enabling direct drop-in use with CLI and IDE platforms including Claude Code, Qwen Code, Cline, Kilo, Trae, and others, without requiring per-platform fine-tuning.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Alibaba | Anthropic | Anthropic |
| Release Date | February 3, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 262K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.50 Best Input Pricing | $4 | $2 |
| Output Pricing | $1.20 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 9.2 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 36.2 | - | - |
| Agentic Index | 0.9 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Alibaba
Other models by Alibaba
Qwen3.8 Max
Qwen3.8 2.4T A95B
Qwen3.8-Flash-Next
Qwen3.8 27B
Top AI Models
Leading alternatives by intelligence score
