# MiniCPM-V 4.7 35B-A3B (slim)

> MiniCPM-V 4.7 35B-A3B is OpenBMB's open-weight vision-language model, a 35B mixture of experts with about 3B active per token that reads images and video, calls tools, and ships in BF16, FP8, 4-bit and GGUF builds under Apache 2.0.

- Full: https://www.anchorterminal.com/tools/minicpm-v-4-7-35b-a3b.md (~10,400 tokens) · this version ~1,830 tokens · JSON https://www.anchorterminal.com/tools/minicpm-v-4-7-35b-a3b.json · canonical https://www.anchorterminal.com/tools/minicpm-v-4-7-35b-a3b
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-10 · pricing https://www.anchorterminal.com/tools/minicpm-v-4-7-35b-a3b#pricing · reviews https://www.anchorterminal.com/tools/minicpm-v-4-7-35b-a3b#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 61/100 · rank #69 of 135 open-weight models · #58 in Open-weight models · not agent-ready · confidence low**


Assessment: A vision-language cousin for images and video that fits a 24 to 32 GB machine from OpenBMB's own GGUF, under plain Apache 2.0, with a publisher that answers. The card says little about limits, training data or how tool calls come back, so test the tool format in your runtime before relying on it.

## Facts

- Kind: Open-weight model · vendor: OpenBMB · category: Open-weight models · legal entity: OpenBMB (named in the GitHub LICENSE's copyright line) · provenance 58/100
- Packages: huggingface `openbmb/MiniCPM-V-4.7-35B-A3B`, huggingface `openbmb/MiniCPM-V-4.7-35B-A3B-gguf`, huggingface `openbmb/MiniCPM-V-4.7-35B-A3B-FP8`, huggingface `openbmb/MiniCPM-V-4.7-35B-A3B-AWQ`, huggingface `openbmb/MiniCPM-V-4.7-35B-A3B-GPTQ`, huggingface `openbmb/MiniCPM-V-4.7-35B-A3B-int4`, ollama `openbmb/minicpm-v4.7:35b-a3b`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0
- Benchmark probes: not run yet, so Performance is pending
- Surface graded: The open weights at huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B and OpenBMB's own GGUF, FP8, AWQ, GPTQ and int4 repositories, read from the model card, the files and their metadata, the chat template, OpenBMB's GitHub README and cookbook and the runtimes' own code on 10 November 2026. Nothing was downloaded or run
- Model: 35,212,875,824 parameters in BF16 per the safetensors metadata, the minicpmv4_7 architecture (MiniCPMV4_7ForConditionalGeneration), a qwen3_5_moe_text language model of 40 layers with 256 experts and 8 routed per token, and a 27-layer vision encoder that compresses visual tokens early, at 4x or 16x. Single images, several images and video in, text out
- Files: Sixteen safetensors shards in BF16 (70.4 GB), a chat_template.jinja, two evaluation-result files and no Python files. No pickle files. Hugging Face lists a SHA-256 for every large file. No LICENSE file in the repository; the licence is in the card's metadata and in OpenBMB's GitHub repository
- Licence: Apache 2.0 in the card's metadata and as the LICENSE file of github.com/OpenBMB/MiniCPM-V. Neither the card nor the GitHub README adds a use policy for this model
- Runtimes: Model code in Transformers from 5.18.0 (still in 5.22.2), vLLM 0.33.0 and SGLang 0.5.23, a guide and the projector type in llama.cpp (b11963 and b12242), and an openbmb/minicpm-v4.7 entry in OpenBMB's namespace on Ollama. OpenBMB's cookbook still says SGLang needs a fork and Ollama waits on its bundled llama.cpp, which the releases have since caught up with. LM Studio's catalogue has no entry
- Context: 262,144 tokens native (max_position_embeddings in config.json, rope_type default). The card doesn't state a context length
- Tools and reasoning: The card shows a tool-calling request through Transformers' OpenAI-compatible server, with tools declared in the OpenAI tools array. The template declares tools in `<tools>` and expects `<function=...>` blocks in `<tool_call>`, several per turn. The card doesn't show a call coming back or name a parser. Thinking is off by default since 19 October 2026 and switches with `enable_thinking`
- Hardware: 70.4 GB in BF16, 38.0 GB in FP8 and 22.7 GB in each of the AWQ, GPTQ and int4 builds. OpenBMB's GGUF files run from 14.5 GB (IQ3_XS) to 69.4 GB (F16), with Q4_K_M at 21.2 GB and a vision projector of 0.9 or 1.1 GB beside them
- Training data: Not described. The card covers the architecture and the on-device techniques and thanks Qwen3.5, Qwen3.6 and SigLIP2, and names no datasets, sizes or knowledge cut-off

- Scores: Capability pending, Reach 90, Risk 55, Agent interface 70, Documentation and upkeep 68 · total over the 7 assessed categories
- Why:  · Reach, Memory 25, since OpenBMB's IQ4_XS GGUF with its vision projector is 19.6 GB and needs about 23.6 GB with the fifth added, which fits 24 GB… · Risk, File safety 25, safetensors and GGUF only, no Python in the main repository and no remote code needed. · Agent interface, Chat template 20, a Jinja file in the repository. · Documentation and upkeep, Model card 8, it covers intended use, architecture and parameters but not limitations, context length, languages or a knowledge cut-off.
- Sources: 29, open questions: 8, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/minicpm-v-4-7-35b-a3b.json
- Its vendor verifies the listing with `verify_listing` or `POST https://www.anchorterminal.com/api/v1/verify` (the full twin says how); no effect on the grade.
- Support: Discord https://discord.gg/3cGQn9b3YM; https://github.com/OpenBMB/MiniCPM-V/issues; https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B/discussions; more in the full twin

## Before you call it

1. Start it with `llama-server -hf openbmb/MiniCPM-V-4.7-35B-A3B-gguf:Q4_K_M --jinja`, or `ollama run openbmb/minicpm-v4.7:35b-a3b`
2. For image input in llama.cpp add `mmproj-model-f16.gguf`, or the 4x projector for more visual detail at a higher cost
3. The GGUF files at the repository root have thinking off. Use the files in `thinking/` when you want the trace
4. With Transformers pass `enable_thinking` to `apply_chat_template`, and pass `downsample_mode` to both `apply_chat_template` and `generate`
5. Use temperature 0.7, top_p 0.95 and top_k 100 in instruct mode, as the card recommends
6. Check tool calls in your runtime first, since the card names no parser. The template uses the Qwen3.5 function-block format
7. Validate JSON output yourself, since structured output isn't documented

## Connect

```bash
ollama run openbmb/minicpm-v4.7:35b-a3b
# or, with llama.cpp and OpenBMB's own GGUF:
llama-server -hf openbmb/MiniCPM-V-4.7-35B-A3B-gguf:Q4_K_M --jinja
# or, with Transformers' server, as the card gives it:
pip install "transformers[serving]>=5.18.0"
transformers serve openbmb/MiniCPM-V-4.7-35B-A3B --port 8000 --host 0.0.0.0
```

```bash
curl http://localhost:11434/api/chat \
  -d '{"model": "openbmb/minicpm-v4.7:35b-a3b", "messages": [{"role": "user", "content": "Hello!"}]}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 75.4 | 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 | 73.6 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 12B IT | BB | 72.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.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 |


## Panel reviews (0, desk reviews from public material, no calls made)