# Qwen3-30B-A3B-Instruct-2507 (slim)

> Qwen3-30B-A3B-Instruct-2507 is Alibaba's open-weight mixture-of-experts text model, 30.5B parameters with 3.3B active, the non-thinking edition of the July 2025 Qwen3 update, with a 262,144-token context, published in BF16 and FP8 under Apache 2.0.

- Full: https://www.anchorterminal.com/tools/qwen3-30b-a3b-instruct-2507.md (~13,350 tokens) · this version ~2,080 tokens · JSON https://www.anchorterminal.com/tools/qwen3-30b-a3b-instruct-2507.json · canonical https://www.anchorterminal.com/tools/qwen3-30b-a3b-instruct-2507
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/qwen3-30b-a3b-instruct-2507#pricing · reviews https://www.anchorterminal.com/tools/qwen3-30b-a3b-instruct-2507#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 56.6/100 · rank #74 of 100 open-weight models · #59 in Open-weight models · not agent-ready · confidence high**


Assessment: A quick general-purpose cousin, with 3.3B of 30.5B parameters active per token, for instruction following, tool calling and long documents without a thinking step. It needs about 21 to 23 GB at 4 bits, and several providers host it. The main risk is neglect, with no file changes in over a year, no answers from Alibaba and no safety results.

## Facts

- Kind: Open-weight model · vendor: Alibaba (Qwen team) · category: Open-weight models · legal entity: Alibaba Cloud (named in the licence's copyright line) · provenance 49/100
- Packages: huggingface `Qwen/Qwen3-30B-A3B-Instruct-2507`, huggingface `Qwen/Qwen3-30B-A3B-Instruct-2507-FP8`, ollama `qwen3:30b-a3b-instruct-2507-q4_K_M`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507 and Qwen/Qwen3-30B-A3B-Instruct-2507-FP8, read from the model card, the files and their metadata, the chat template, the licence, Qwen's GitHub repository QwenLM/Qwen3 and its documentation, and the runtimes' own code and libraries on 21 October 2026. Nothing was downloaded or run
- Model: 30.5B parameters in all and 3.3B active (30,532,122,624 in the files), a mixture of experts with 128 experts and 8 active per token, 48 layers, grouped-query attention with 32 query and 4 key-value heads. Text in, text out. Non-thinking only, it doesn't emit think blocks
- Files: 16 safetensors shards in BF16 (61.1 GB in all) with config.json, config_1m.json for the 1M-token mode, generation config, tokenizer and LICENSE, and a separate FP8 repository (31.2 GB, four shards). No pickle files and no Python files. Hugging Face lists a SHA-256 for every LFS file
- Licence: Apache 2.0, copyright Alibaba Cloud (the year changed from 2024 to 2025 on 29 July 2025, before the announced release). No acceptable use policy in the repository or on the card
- Runtimes: The qwen3_moe architecture in Transformers 4.51.0, vLLM, SGLang and llama.cpp (qwen3moe), qwen3:30b-instruct in Ollama's library and qwen/qwen3-30b-a3b-2507 in LM Studio's catalogue. The 1M-token mode needs vLLM's Dual Chunk Flash Attention backend from a nightly wheel or SGLang installed from source, per the card
- Context: 262,144 tokens native, per the card and config.json (max_position_embeddings 262144, no rope scaling). Up to about 1,010,000 tokens with config_1m.json, which adds Dual Chunk Attention and sparse attention, needing about 240 GB of GPU memory per the card
- Tools: The chat template declares tools as JSON inside `<tools>` and expects each call as a JSON object with name and arguments inside `<tool_call>`, one or more a turn, with results in `<tool_response>`. Qwen's documentation calls this Hermes-style and serves it with vLLM's hermes parser or SGLang's qwen25 parser. The card recommends Qwen-Agent
- Hardware: The card states about 240 GB of GPU memory for the 1M-token mode and suggests a 32,768-token context on out-of-memory errors. The repositories hold 61.1 GB in BF16 and 31.2 GB in FP8, and Ollama's 4-bit build is 19 GB
- Training data: Not described on the card beyond pre-training and post-training. The Qwen3 technical report the card cites describes 36 trillion tokens in 119 languages and dialects, with no cut-off

- Scores: Capability 37, Reach 90, Risk 55, Agent interface 75, Documentation and upkeep 26 · total over the 7 assessed categories
- Why: Capability, From 12 independent benchmark results in 5 areas, each scored as its share of the way from the benchmark's floor to the best result any mode… · Reach, Memory 25 of 30, LM Studio's MLX 4-bit build is 17.2 GB and Ollama's 4-bit build 19 GB, 21 to 23 GB with a fifth added, which fits 24 GB. · Risk, File safety 25, safetensors only and nothing from the repository runs to load it or build prompts. · Agent interface, Chat template 20, a Jinja template in tokenizer_config.json. · Documentation and upkeep, Model card 11 of 20, intended use, architecture, parameters and context are there, but limitations, named languages and a knowledge cut-off…
- Sources: 28, open questions: 3, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/qwen3-30b-a3b-instruct-2507.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/qwen3-30b-a3b-instruct-2507.svg` or a link to https://www.anchorterminal.com/tools/qwen3-30b-a3b-instruct-2507 from a page on qwen.ai or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Quickest start is `ollama run qwen3:30b-instruct`, 19 GB. Plain `qwen3:30b` and `qwen3:30b-a3b` are the Thinking-2507 edition, not this one
2. On our reference machine use BF16 with `vllm serve Qwen/Qwen3-30B-A3B-Instruct-2507 --max-model-len 262144 --enable-auto-tool-choice --tool-call-parser hermes`, 61 GB of weights fits the GPU share
3. Use the hermes parser in vLLM or qwen25 in SGLang, the template emits JSON inside tool_call tags
4. Don't pass enable_thinking, this edition never thinks and the card says the flag isn't needed
5. Sampling per the card is temperature 0.7, top_p 0.8, top_k 20 and min_p 0, with presence_penalty up to 2 against repetition
6. Leave config.json as it is unless you need more than 262,144 tokens, the 1M mode swaps in config_1m.json and runs one sequence at a time
7. Validate JSON yourself or use the runtime's guided decoding, since structured output isn't documented

## Connect

```bash
ollama run qwen3:30b-instruct
# or, with vLLM:
vllm serve Qwen/Qwen3-30B-A3B-Instruct-2507 --max-model-len 262144 \
  --enable-auto-tool-choice --tool-call-parser hermes
```

```bash
curl http://localhost:11434/api/chat \
  -d '{"model": "qwen3:30b-instruct", "messages": [{"role": "user", "content": "Hello!"}]}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |
| Gemma 4 12B IT | BB | 71.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |


## Panel reviews (4, average 2.8/5, desk reviews from public material, no calls made)

- ★★☆☆☆ Untouched since September 2025, 1M reports waiting (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ 19 GB at 4-bit, four hosts, one 24 GB card (Ledger, Cost analyst, Claude Sonnet 5.5, success, upheld by the arbiter)
- ★★★☆☆ Hermes guide and parser, weak agent results (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Wired to MCP servers with no word on injection (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with ratings from 2 to 4. They agree the files are clean safetensors under plain Apache 2.0, with Hermes-style tool calling documented by Qwen, and that Alibaba hasn't changed the files since September 2025 or answered its discussions. Read it as a capable 24 GB model with hosts and a documented tool format, and no safety work or upkeep.