Alibaba Qwen 3 family
https://qwenlm.github.io/blog/qwen3/
Alibaba Cloud's Qwen 3 family (2025-04) — open-weights frontier models from 0.5B to 235B-A22B MoE. Apache 2.0 (most sizes) — broadest open-weights family. Includes Qwen 3 32B dense, Qwen 3 30B-A3B MoE, Qwen 3 235B-A22B MoE. Qwen-Coder, Qwen-VL (vision), Qwen-Audio companion variants. Dominant base model for many Chinese AI startups and a top OSS substrate globally.
At a glance
- Type
- Open-weights frontier model family (Qwen 3 dense + MoE; 0.5B–235B)
- Tier
- T1
- Created
- 2023-08 (Qwen 1); 2024-02 (Qwen 1.5); 2024-09 (Qwen 2.5); 2025-04 (Qwen 3)
- Latest release
- Qwen 3 family (2025-04); Qwen 3-Max preview (2025-Q3)
- License
- Apache 2.0 (most Qwen 3 sizes); some larger MoE variants under Qwen License
- Pricing
- Free open-weights (Apache 2.0); Alibaba Cloud API pay-per-token ($0.10-$2 per 1M depending on tier); Together / Fireworks hosted
- Funding
- Alibaba (BABA) public; ~$200B market cap
Taxonomy
- storage
- parametric
- retrieval
- parametric-recall
- persistence
- parametric-permanent
- update
- read-only
- unit
- weight
- governance
- opaque
- conflict
- n/a
When to use
Optimised for: multilingual (119 languages including Asian + minor European); cost-efficient inference; permissive Apache 2.0 OSS
Anti-fit: China data-residency for regulated Western workloads; less mainstream Western enterprise adoption; Qwen-Max not open-weights
Pros & cons
Pros
Broadest open-weights frontier family (0.5B to 235B); Apache 2.0 (most permissive); 119-language coverage; strong reasoning at small / medium sizes; #1-ranked OSS family on HF for many months 2024-25.
Cons
China-based — regulatory risk for some Western enterprises; multimodal less polished than Gemini 3; Qwen-Max tier (rumored 1T+) is closed.
Claims & capabilities
Qwen 3 family Apache 2.0; 0.5B to 235B-A22B MoE; 119 languages; thinking/non-thinking dual mode; matches DeepSeek R1 on reasoning at smaller sizes
Technical surface
- API surface
- OpenAI-compatible REST (Alibaba Cloud Model Studio); HuggingFace Transformers; vLLM; llama.cpp
- Backend storage
- not applicable — substrate foundation model
- Deployment
- Self-hostable (Apache 2.0 weights) + Alibaba Cloud Model Studio + Together AI + Fireworks + HuggingFace inference + Ollama
- Embedding model
- not applicable — not a memory product
- Multi-tenancy
- not applicable — substrate foundation model
- MCP
- not applicable — substrate foundation model
- A2A
- not applicable — substrate foundation model
- OpenTelemetry
- not applicable — substrate foundation model
Similar systems
Other foundation models (substrate reference) in the catalog, ranked by inbound references.
- Anthropic Claude (foundation models) T1
Claude family of frontier foundation models — Sonnet 4 (workhorse), Opus 4.5 (frontier capability), Haiku 4.5 (fast/cheap). Frontier model used as substrate by Claude Code, Anthropic Memory tool, Cline, Aider, Continue.dev, Cursor, Goose, and dozens of other catalog entries. Memory in Claude is parametric (in weights) — discrete episodic memory lives outside the model in the Claude Memory tool (see Platform-provider memory section).
- OpenAI GPT family (GPT-5 / GPT-4o / o3 / o4) T1
OpenAI's foundation model family — GPT-5 (2025-08 flagship), GPT-4o (multimodal workhorse), o3/o4 reasoning series. Substrate for ChatGPT, the OpenAI API, Microsoft Copilot, Azure OpenAI, and many downstream agents in this catalog (Devin, Codex CLI, Operator, Cursor mode). Parametric memory in weights; discrete user memory ships in ChatGPT Memory feature (separate row).
- Google Gemini 3 family T1
Google DeepMind's Gemini 3 frontier model family — Gemini 3 Pro (top tier), 3 Flash (fast), 3 Nano (on-device). Substrate for Google Workspace AI, Vertex AI Agent Platform, Gemini Code Assist, Project Mariner, and Project Astra. Multimodal-native; 1M-10M context window. Largest context window of any frontier model. Used by Gemini Enterprise Agent Platform Memory Bank (separate row).
- Mistral Large 2 / Mixtral family T1
French frontier-model lab (Paris). Family includes Mistral Large 2 (123B dense, Jul-2024), Mixtral 8x22B (open-weights MoE), Mistral Small 3 (24B, open-weights Jan-2025), Codestral (code specialist). EU-headquartered alternative to US/China model labs; GDPR-friendly data residency. Distinct from the Mistral Le Chat memory feature (already in catalog) — this row covers the model family itself as substrate.
- Meta Llama 4 family T1
Meta's Llama 4 family (released 2025-04) — Scout (109B MoE, 17B active), Maverick (400B MoE, 17B active), Behemoth (2T MoE, 288B active; preview). First Llama generation to use mixture-of-experts and to be multimodal-native. Substrate for many OSS agents (Open-Interpreter, Ollama-served local agents, vLLM-served enterprise inference). Llama community license (commercial use permitted under 700M MAU threshold).
- xAI Grok 4 T1
Elon Musk's xAI frontier model — Grok 4 launched 2025-07. Trained on Colossus (100k+ GPU supercluster in Memphis). Deeply integrated with X/Twitter (real-time tweet data access). Grok 4 Heavy variant for premium tier. No frontier-tier open-weights — Grok 1 (314B MoE) is the only weight release.