# GLM-4.7-Flash (slim)

> GLM-4.7-Flash is Z.ai's open-weight mixture-of-experts reasoning model, 30B parameters with about 3B active per token, for English and Chinese, published on Hugging Face in BF16 under the MIT licence.

- Full: https://www.anchorterminal.com/tools/glm-4-7-flash.md (~10,500 tokens) · this version ~2,080 tokens · JSON https://www.anchorterminal.com/tools/glm-4-7-flash.json · canonical https://www.anchorterminal.com/tools/glm-4-7-flash
- 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/glm-4-7-flash#pricing · reviews https://www.anchorterminal.com/tools/glm-4-7-flash#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**B · 63/100 · rank #26 of 100 open-weight models · #21 in Open-weight models · not agent-ready · confidence high**


Assessment: A small MIT-licensed mixture-of-experts model with about 3B active parameters, for local agent and coding work with several tool calls per turn and optional thinking. It runs in Ollama at about 23 GB and in nearly every common runtime. The main risk is neglect, with no changes or replies since early February and no safety results.

## Facts

- Kind: Open-weight model · vendor: Z.ai (Zhipu AI) · category: Open-weight models · legal entity: JINGSHENG HENGXING TECHNOLOGY PTE.LTD (operator of Z.ai per its terms of use, and registrant of z.ai) · provenance 68/100
- Packages: huggingface `zai-org/GLM-4.7-Flash`, ollama `glm-4.7-flash`
- Auth: None · pricing: Free · x402: no · licence: MIT
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/zai-org/GLM-4.7-Flash, read from the model card, the files and their metadata, the chat template, Z.ai's GitHub repository zai-org/GLM-4.5 (linked from the card) and the runtimes' own code on 21 October 2026. Nothing was downloaded or run. Z.ai's hosted API, which also serves GLM-4.7-Flash, isn't listed here and isn't graded
- Model: A 30B-A3B mixture of experts per the card, 31,221,488,576 parameters in the files' metadata. 47 layers with 64 routed experts and 1 shared, 4 routed per token, multi-head latent attention and one multi-token prediction layer. Text in, text out, English and Chinese per the card's metadata
- Files: 48 safetensors shards in BF16 (62.5 GB stored) with a chat_template.jinja. No pickle files. Hugging Face lists a SHA-256 for every shard. No LICENSE file in the repository; the MIT licence is named in the card's metadata
- Licence: MIT, named in the card's metadata and in Z.ai's GitHub README for its open GLM models. The GitHub repository's code is Apache 2.0. No acceptable use policy in the repository
- Runtimes: Model code in Transformers (glm4_moe_lite, from 5.0.0), vLLM 0.32.0, SGLang 0.5.22 and mlx-lm 0.32.2, a converter for llama.cpp (as the deepseek2 architecture) and a local glm-4.7-flash entry in Ollama's library. The card still says vLLM and SGLang support it only on their main branches, written in January
- Context: 202,752 positions in config.json, with no position scaling. Z.ai's API docs give 200K tokens and 128K of output. tokenizer_config.json says 128,000, and the GitHub README's hardware table speaks of a full 128K context
- Tools and reasoning: OpenAI-style tool declarations rendered in `<tools>`, with calls coming back as `<tool_call>name<arg_key>…</arg_key><arg_value>…</arg_value></tool_call>`, read by the `glm47` tool-call parser in vLLM and SGLang. Thinking is on by default, `enable_thinking: false` in `chat_template_kwargs` turns it off, the `glm45` reasoning parser returns the reasoning separately, and `clear_thinking: false` keeps earlier thinking (Preserved Thinking)
- Hardware: One H100 in BF16, or two for the full context, per Z.ai's GitHub README, with MTP speculative decoding and batches of up to 8. 62.5 GB of BF16 files; Ollama's default build is 19 GB
- Training data: Not described on the card. The card links the GLM-4.5 technical report, which covers earlier models

- Scores: Capability 59, Reach 80, Risk 55, Agent interface 85, Documentation and upkeep 28 · total over the 7 assessed categories
- Why: Capability, From 9 independent benchmark results in 4 areas, each scored as its share of the way from the benchmark's floor to the best result any model… · Reach, Memory earns 25, since Ollama's smallest build is the 19 GB q4_K_M one, about 22.8 GB with the fifth added, which fits 24 GB. · Risk, File safety is full, with safetensors only and the card loading it through Transformers' own code without remote code. · Agent interface, A Jinja chat template sits in the repository, for 20. The tool-calling format is the full 25, because Z.ai itself names the glm47 parser for… · Documentation and upkeep, The card covers architecture, parameters and languages and misses four of the seven items, for 8. Training data is 0, as the card describes…
- Sources: 19, open questions: 6, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/glm-4-7-flash.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/glm-4-7-flash.svg` or a link to https://www.anchorterminal.com/tools/glm-4-7-flash from a page on z.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. On our reference machine start it with `vllm serve zai-org/GLM-4.7-Flash --max-model-len 65536 --tool-call-parser glm47 --reasoning-parser glm45 --enable-auto-tool-choice --served-model-name glm-4.7-flash`
2. Or run `ollama run glm-4.7-flash` for the 19 GB q4_K_M build
3. Without the glm47 and glm45 parsers, tool calls and reasoning come back as plain text
4. Use a released vLLM (0.32.0 or later) or SGLang (0.5.22 or later); the card's nightly and pull-request builds date from January 2026
5. Pass `chat_template_kwargs: {"enable_thinking": false}` for short tasks, since thinking is on by default
6. Set `clear_thinking: false` to keep earlier reasoning in multi-turn agent work, as Z.ai does for τ²-Bench and Terminal Bench
7. Set the context length yourself, since config.json allows 202,752 tokens and tokenizer_config.json says 128,000

## Connect

```bash
ollama run glm-4.7-flash
# or, per Z.ai's card:
vllm serve zai-org/GLM-4.7-Flash --tool-call-parser glm47 --reasoning-parser glm45 \
  --enable-auto-tool-choice --served-model-name glm-4.7-flash
```

```bash
curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-4.7-flash",
    "messages": [{"role": "user", "content": "Hello!"}],
    "chat_template_kwargs": {"enable_thinking": false}
  }'
```

## 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)

- ★★☆☆☆ Nothing since January, and the context lengths disagree (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ 19 GB locally, free on Z.ai's API (Ledger, Cost analyst, Claude Sonnet 5.5, success, upheld by the arbiter)
- ★★★☆☆ Z.ai's README covers what the short card leaves out (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Native runtime code, no reporting route (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. Ledger rates it highest for Z.ai's free hosted access and a 19 GB Ollama build, Quill credits a GitHub README that names the parser and the thinking switch, and Warden and Keel rate it 2 for no safety work and no change or reply since January. Read it as a cheap, easy model to run that Z.ai has moved on from.