Qwen3

Type: AI model family

Qwen3 is Alibaba’s open-weight model family released in dense and mixture-of-experts sizes from 0.6B to 235B-A22B, all Apache 2.0 licensed, with seamless switching between reasoning and non-thinking modes. It surpassed Qwen2.5 on math, code, and reasoning. GROUNDING tracks the Qwen3 family’s variants, benchmarks, and ecosystem adoption.

Recent Updates

• 2026-07-03: Program-as-Weights proposes compiling fuzzy functions into locally executable neural artifacts (cs.CL updates on arXiv.org) · arxiv.org — Qwen3-32B
• 2026-07-03: MMIR-TCM combines memory-augmented segmentation, Qwen3-VL, and RAG for TCM diagnosis (cs.AI updates on arXiv.org) · arxiv.org — RAG LLM Evals Qwen3-VL · GPT-4o · Gemini 2.5 Flash Memory-SAM
• 2026-07-03: Program-as-Weights compiles natural-language specs into small local neural functions (Hacker News) · arxiv.org — arXiv · arXivLabs · Connected Papers · Litmaps · scite · alphaXiv · CatalyzeX · DagsHub · Gotit.pub · Hugging Face · ScienceCast · Influence Flower · CORE Recommender · IArxiv · Qwen3-32B
• 2026-07-05: Microsoft-linked code search SLM reportedly appeared on Hugging Face and was later removed (AI Projects) · t.me — Codebase Indexing Tool Use LLM Evals Microsoft · Hugging Face
• 2026-07-06: LLamaFactory Enters GitHub Top 10 AI Rankings with Unified Fine-Tuning Framework (GitHub AI Ranking Changes (Top 10)) · github.com — Open Source LLMs Google · Alibaba · Amazon · NVIDIA · DeepSeek Apoidea Group · LLaMA · LLaVA · Mistral Mixtral-MoE · Qwen3-VL Qwen2.5-Omni · DeepSeek-R1-Distill-Qwen-7B · Gemma · GLM GLM-Z1 GLM-4.1V-9B-Thinking · Phi InternVL3 InternS1-mini Kimi-VL · GPT-OSS
• 2026-07-06: Ludwig: Declarative Framework for LLM Fine-Tuning and Multimodal AI Model Training (GitHub AI Ranking Changes (Top 10)) · github.com — Linux Foundation · Hugging Face · Meta · Llama 3.1 Qwen2-VL InternVL · BERT · RoBERTa · ModernBERT · Mamba-2 Jamba · DINOv2 ConvNeXt EfficientNet · ViT CAFormer ConvFormer PoolFormer · PatchTST N-BEATS TabNet TabPFN v2 UNet SegFormer
• 2026-07-06: ms-swift ranks #4 on GitHub AI with unified fine-tuning support for 600+ text and 400+ multimodal models (GitHub AI Ranking Changes (Top 10)) · github.com — Open Source LLMs ModelScope · Qwen3.5 · Qwen3.6 · DeepSeek R1 · DeepSeek V4 DeepSeek-VL2 Llama4 InternLM3 GLM4.5 GLM4.5-V · GLM-5.1 · Mistral · Qwen3-VL Qwen3-Omni InternVL3.5 Ovis2.5 · Gemma4 · LLaVA Phi4 MiniCPM-V-4
• 2026-07-07: Classification-head fine-tuning makes tiny Qwen3 models stronger on multiple-choice benchmarks (cs.LG updates on arXiv.org) · arxiv.org — arXiv · Qwen3-0.6B · Qwen3-1.7B · GPT-3 PaLM · GPT-4
• 2026-07-07: Self-distillation can hurt strong thinking models on long reasoning traces (cs.AI updates on arXiv.org) · arxiv.org — LLM Evals Olmo
• 2026-07-08: Self-play can exploit reference-free LLM judges by optimizing plausibility over correctness (cs.LG updates on arXiv.org) · arxiv.org — LLM Evals Qwen · LLaMA · Gemma
• 2026-07-09: Agon: cross-model RL that grades reasoning by competition (cs.CL updates on arXiv.org) · arxiv.org — LLM Evals Qwen3.5 · Gemma 4
• 2026-07-09: DeLS-Spec proposes a decoupled long-short context method for speculative decoding (cs.CL updates on arXiv.org) · arxiv.org — DFlash Domino DSpark
• 2026-07-10: Auditing how LLM judges shift when the evaluator changes (cs.CL updates on arXiv.org) · arxiv.org — LLM Evals MiniMax · Qwen MiniMax M2 · MiniMax-M2.7
• 2026-07-13: Signed Symmetric Quantization for Few-Bit Integers (cs.LG updates on arXiv.org) · arxiv.org — AMD · Qwen3.5 · Llama3
• 2026-07-14: Budgeted placement of strong correctors in weak multi-agent swarms (cs.CL updates on arXiv.org) · arxiv.org — Agents LLM Evals
• 2026-07-14: Continual fact writing into model weights appears fragile compared with context (cs.CL updates on arXiv.org) · arxiv.org — Context Engineering arXiv
• 2026-07-14: Budgeted placement of strong correctors in a weak multi-agent swarm (cs.AI updates on arXiv.org) · arxiv.org — Agents
• 2026-07-16: Entity familiarity probes and refusal steering in instruction-tuned models (cs.CL updates on arXiv.org) · arxiv.org — LLM Evals arXiv · Wikipedia Grzegorz Brzezinka · Bielik PLLuM · Gemma 4 · Gemma 4 12B
• 2026-07-20: On-Policy Delta Distillation focuses training on reasoning changes (alphaXiv) · x.com — alphaXiv · Gemma4
• 2026-07-20: On-Policy Delta Distillation for Reasoning Models (alphaXiv) — NAVER AI Lab Byeongho Heo Jaehui Hwang Sangdoo Yun Dongyoon Han Gemma 4
• 2026-07-20: Better Starts, Better Ends: Bootstrapped Iterative Self-Reasoning Distillation for Compressed Reasoning (cs.CL updates on arXiv.org) · arxiv.org — LLM Evals Qwen3-8B
• 2026-07-21: Masked diffusion world models for agentic RL across open-source and frontier backbones (cs.AI updates on arXiv.org) · arxiv.org — Agents Tool Use Context Engineering LLM Evals LFM2.5 Mistral
• 2026-07-22: PEARL adds solver-in-the-loop optimization modeling from natural language (cs.AI updates on arXiv.org) · arxiv.org — Agents Tool Use Context Engineering arXiv · DeepSeek-V3.2
• 2026-07-22: Depthwise Convolutions Improve Qwen3 Block Performance with Minimal Parameter Cost (cs.CL updates on arXiv.org) · arxiv.org
• 2026-07-24: Learn2Zinc fine-tunes small language models for MiniZinc text-to-model translation (cs.CL updates on arXiv.org) · arxiv.org — arXiv · LLaMA · Gemma · GPT-OSS

FAQ

What is Qwen3?

Qwen3 is Alibaba’s open-weight model family released in dense and mixture-of-experts sizes from 0.6B to 235B-A22B, all Apache 2.0 licensed, with seamless switching between reasoning and non-thinking modes. It surpassed Qwen2.5 on math, code, and reasoning. GROUNDING tracks the Qwen3 family’s variants, benchmarks, and ecosystem adoption.

What does this page track?

Dated radar mentions, source links, related concepts, and builder-relevant context for Qwen3, collected automatically by GROUNDING.

When was Qwen3 last mentioned?

Qwen3 was most recently mentioned in a radar update dated 2026-07-24.

Category: Text / Language Models