# MiniCPM-o-4_5 by OpenBMB: Open-Weight Model Source: https://savrn.com/models/minicpm-o-4-5 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve MiniCPM-o-4_5 (9.4B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 18.7 GB | 22.5 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 9.4 GB | 11.2 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 4.7 GB | 5.6 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 25, 2026. [MiniCPM-o-4_5 on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/minicpm-o-4-5/gpus) ## SAVRN's Notes on MiniCPM-o-4_5 9.4B parameters, 20.0 GB of weights across eight files, 22.5 GB at 16-bit and 5.6 GB at 4-bit: an any-to-any model that sees, hears and speaks, servable on one MI300X at $1.85 an hour — or, per the card's own pitch, on a phone. The 40,960-token context is modest against text-only rivals, but full-duplex streaming across vision and speech is the point, not raw context. Apache 2.0 permits commercial use with notices kept. The page lists an official AWQ build, MiniCPM-o-4_5-awq, quantized from this checkpoint, if you want a packaged 4-bit path rather than rolling your own. The 8-file layout reflects the multi-tower design — separate weights for vision and audio beside the language model — so serving it means orchestrating modules, not loading one blob. No third-party evaluations are on file; the Gemini-2.5-Flash-level claim is the publisher's. SAVRN Research ## Model Card By OpenBMB, published under apache-2.0, revision 503e754207c9. A Gemini 2.5 Flash Level MLLM for Vision, Speech, and Full-Duplex Mulitmodal Live Streaming on Your Phone [GitHub](https://github.com/OpenBMB/MiniCPM-o) | [CookBook](https://github.com/OpenSQZ/MiniCPM-V-CookBook) | [Omni-modal Demo](https://minicpmo45.modelbest.cn/) | [Vision-Language Demo](http://211.93.21.133:18121/) [WeChat](https://github.com/OpenBMB/MiniCPM-o/blob/main/docs/wechat.md) | [Discord](https://discord.gg/N2RnxGdJ) | [MiniCPM Wiki(Chinese)](https://modelbest.feishu.cn/wiki/UtWxwcERfiRIpIkBOjuc3h9tn1D) | CaseBook([Audio](https://openbmb.github.io/minicpm-o-4_5/), [Omni Full-Duplex](https://openbmb.github.io/minicpm-o-4_5-omni/)) ### News [!NOTE] [2026.02.06] We open-sourced a realtime web demo deployable on your own devices like Mac or GPU.Try it now! ### MiniCPM-o 4.5 MiniCPM-o 4.5 is the latest and most capable model in the MiniCPM-o series. The model is built in an end-to-end fashion based on SigLip2, Whisper-medium, CosyVoice2, and Qwen3-8B with a total of 9B parameters. It exhibits a significant performance improvement, and introduces new features for full-duplex multimodal live streaming. Notable features of MiniCPM-o 4.5 include: [Read the full model card (6,159 words)](https://savrn.com/models/minicpm-o-4-5/card) ## Configuration Architecture MiniCPMO Context length (tokens) 40,960 Layers 36 Hidden size 4,096 Feed-forward size 12,288 Attention heads 32 Key/value heads 8 Head dimension 128 Vocabulary size 151,748 RoPE base 1,000,000 Stored precision bfloat16 Model type minicpmo ## Identity and Version Repository openbmb/MiniCPM-o-4_5 Publisher OpenBMB Task Any to any Modality Multimodal Library transformers Parameters 9.4B parameters Languages Not stated by the source Revision 503e754207c94da6bb26850b4469f367c9ea3582 First published 2026-02-03 Last updated 2026-08-18 ## Files and Weights 54 files, 20.0 GB in total. The weights are 8 files totalling 20.0 GB in onnx, pt, safetensors. Weights8 files · 20.0 GB Configuration13 files · 573.8 KB Tokenizer4 files · 15.7 MB Documentation1 file · 85.8 KB Other26 files · 57.3 MB Repository2 files · 3.1 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | assets/token2wav/campplus.onnx | Weights | 28.3 MB | a6ac6a639977 | | assets/token2wav/flow.pt | Weights | 623.5 MB | 15ccff24256f | | assets/token2wav/hift.pt | Weights | 83.4 MB | 3386cc880324 | | assets/token2wav/speech_tokenizer_v2_25hz.onnx | Weights | 496.1 MB | d43342aa1216 | | model-00001-of-00004.safetensors | Weights | 5.3 GB | 30c40b9a1038 | | model-00002-of-00004.safetensors | Weights | 5.3 GB | fe0faef420ac | | model-00003-of-00004.safetensors | Weights | 5.3 GB | 5d0b20153f9b | | model-00004-of-00004.safetensors | Weights | 2.9 GB | f61addf4747c | | added_tokens.json | Configuration | 2.7 KB | — | | assets/token2wav/flow.yaml | Configuration | 1.1 KB | — | | config.json | Configuration | 6.4 KB | — | | configuration_minicpmo.py | Configuration | 10.1 KB | — | | generation_config.json | Configuration | 178 B | — | | model.safetensors.index.json | Configuration | 117.2 KB | — | | modeling_minicpmo.py | Configuration | 214.4 KB | — | | modeling_navit_siglip.py | Configuration | 42.6 KB | — | | preprocessor_config.json | Configuration | 842 B | — | | processing_minicpmo.py | Configuration | 67.3 KB | — | | special_tokens_map.json | Configuration | 12.0 KB | — | | tokenization_minicpmo_fast.py | Configuration | 3.2 KB | — | | utils.py | Configuration | 95.7 KB | — | | README.md | Documentation | 85.8 KB | — | | assets/HT_ref_audio.wav | Other | 192.6 KB | cb8f06ba5080 | | assets/Skiing.mp4 | Other | 8.5 MB | 479ace116d6a | | assets/Trump_WEF_2018_10s.mp3 | Other | 161.1 KB | 4fb796c2bb95 | | assets/audio_cases/assistant_ref.mp4 | Other | 65.5 KB | 7e4a56e44187 | | assets/audio_cases/assistant_response.mp4 | Other | 269.5 KB | d46268e3beb7 | | assets/audio_cases/elon_musk__000_assistant_audio.wav | Other | 1.8 MB | ff6dced2e686 | | assets/audio_cases/elon_musk__system_ref_audio.wav | Other | 539.0 KB | 2c4109b2d685 | | assets/audio_cases/elon_musk_ref.mp4 | Other | 165.2 KB | 206c48ec4b08 | | assets/audio_cases/elon_musk_response.mp4 | Other | 388.5 KB | d83c3a977fd4 | | assets/audio_cases/hermione__000_assistant_audio.wav | Other | 1.5 MB | cfd18a2ee9e2 | | assets/audio_cases/hermione__system_ref_audio.wav | Other | 197.3 KB | 46bd82796ce5 | | assets/audio_cases/minicpm_assistant__000_assistant_audio.wav | Other | 1.2 MB | fe3b793a6436 | | assets/audio_cases/minicpm_assistant__system_ref_audio.wav | Other | 192.6 KB | ad576b50fd2f | | assets/audio_cases/paimon__000_assistant_audio.wav | Other | 697.0 KB | 9ddd9d835895 | | assets/audio_cases/paimon__system_ref_audio.wav | Other | 479.3 KB | b9625e7115ff | | assets/audio_cases/readme.txt | Other | 57 B | — | | assets/bajie.wav | Other | 636.5 KB | 16aa8ca3da7d | | assets/fossil.png | Other | 465.9 KB | b8b3f1668da6 | | assets/haimianbaobao.wav | Other | 343.1 KB | 27405cf5977f | | assets/highway.png | Other | 840.9 KB | 87c32da6ee77 | | assets/nezha.wav | Other | 457.8 KB | fe5d8932013f | | assets/omni_duplex1.mp4 | Other | 7.3 MB | 31622e1efd9a | | assets/omni_duplex2.mp4 | Other | 29.3 MB | c04eaef27a82 | | assets/sunwukong.wav | Other | 644.6 KB | 3bdb6c175bd3 | | assets/system_ref_audio.wav | Other | 539.0 KB | 2c4109b2d685 | | assets/system_ref_audio_2.wav | Other | 341.2 KB | 4a65d1709923 | | .gitattributes | Repository | 3.1 KB | — | | .gitignore | Repository | 9 B | — | | merges.txt | Tokenizer | 1.4 MB | — | | tokenizer.json | Tokenizer | 11.4 MB | 6d55eb34389b | | tokenizer_config.json | Tokenizer | 101.7 KB | — | | vocab.json | Tokenizer | 2.8 MB | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 20.0 GB [Download from OpenBMB](https://huggingface.co/openbmb/MiniCPM-o-4_5) Released by OpenBMB through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## Built From - Described by [arXiv:2408.01800](https://savrn.com/papers/minicpm-v-a-gpt-4v-level-mllm-on-your-phone) ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 20.0 GB | | 16-bit | 18.7 GB | | 8-bit | 9.4 GB | | 4-bit | 4.7 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Built on This Model - Quantized from[MiniCPM-o-4_5-awq](https://savrn.com/models/minicpm-o-4-5-awq) - Derived from[MiniCPM-o-4_5-awq](https://savrn.com/models/minicpm-o-4-5-awq) - Quantized from[MiniCPM-o-4_5-gguf](https://savrn.com/models/minicpm-o-4-5-gguf) - Derived from[MiniCPM-o-4_5-gguf](https://savrn.com/models/minicpm-o-4-5-gguf) ## Compare MiniCPM-o-4_5 - [gemma-4-12B-it vs MiniCPM-o-4_5](https://savrn.com/models/compare/gemma-4-12b-it-vs-minicpm-o-4-5) - [gemma-4-E4B-it vs MiniCPM-o-4_5](https://savrn.com/models/compare/gemma-4-e4b-it-vs-minicpm-o-4-5) ## Questions About MiniCPM-o-4_5 ### How much GPU memory does MiniCPM-o-4_5 need? About 22.5 GB at 16-bit and 5.6 GB at 4-bit: the weights (9.4B parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run MiniCPM-o-4_5 on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use MiniCPM-o-4_5 commercially? Yes. MiniCPM-o-4_5 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors. ### What is MiniCPM-o-4_5's context length? 40,960 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Any to any ### [MiniCPM-o-4_5-awq](https://savrn.com/models/minicpm-o-4-5-awq) [OpenBMB](https://savrn.com/model-publishers/openbmb) A Gemini 2.5 Flash Level MLLM for Vision, Speech, and Full-Duplex Mulitmodal Live Streaming on | CaseBook(Audio, Omni Full-Duplex) MiniCPM-o 4.5 is the latest and most capable model in the MiniCPM-o series. The model is built in an end-to-end fashion based on SigLip2, Whisper-medium, CosyVoice2, and Qwen3-8B with a total of 9B parameters. It exhibits a significant performance improvement, and introduces new features for full-duplex multimodal live streaming. Notable features of MiniCPM-o 4.5 include: - Leading Visual Capability. MiniCPM-o 4.5 achieves an average score of 77.6 on OpenCompass, a comprehensive evaluation of 8 popular benchmarks. With only 9B parameters, it surpasses widely… Open weights apache-2.0 9.4B parameters 40,960 tokens transformers [View model](https://savrn.com/models/minicpm-o-4-5-awq) Model · Any to any ### [gemma-4-E4B-it-qat-w4a16-ct](https://savrn.com/models/gemma-4-e4b-it-qat-w4a16-ct) [Google](https://savrn.com/model-publishers/google) Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from… Open weights apache-2.0 8.7B parameters 131,072 tokens transformers [View model](https://savrn.com/models/gemma-4-e4b-it-qat-w4a16-ct) Model · Any to any ### [MiniCPM-o-2_6](https://savrn.com/models/minicpm-o-2-6) [OpenBMB](https://savrn.com/model-publishers/openbmb) [2025.06.20] Our official ollama repository is released. Try our latest models with one click！ [2025.03.01] RLAIF-V, which is the alignment technique of MiniCPM-o, is accepted by CVPR 2025！The code, dataset, paper are open-sourced! [2025.01.24] MiniCPM-o 2.6 technical report is released! See Here. [2025.01.19] MiniCPM-o tops GitHub Trending and reaches top-2 on Hugging Face Trending! MiniCPM-o 2.6 is the latest and most capable model in the MiniCPM-o series. The model is built in an end-to-end fashion based on SigLip-400M, Whisper-medium-300M, ChatTTS-200M, and Qwen2.5-7B with a total of 8B parameters. It exhibits a significant performance improvement over MiniCPM-V 2.6, and introduces new… Open weights apache-2.0 8.7B parameters 32,768 tokens transformers [View model](https://savrn.com/models/minicpm-o-2-6) Model · Any to any ### [Qwen2.5-Omni-7B](https://savrn.com/models/qwen2-5-omni-7b) [Qwen](https://savrn.com/model-publishers/qwen) Qwen2.5-Omni is an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. We conducted a comprehensive evaluation of Qwen2.5-Omni, which demonstrates strong performance across all modalities when compared to similarly sized single-modality models and closed-source models like Qwen2.5-VL-7B, Qwen2-Audio, and Gemini-1.5-pro. In tasks requiring the integration of multiple modalities, such as OmniBench, Qwen2.5-Omni achieves state-of-the-art performance. Furthermore, in single-modality tasks, it excels in areas including speech recognition (Common… Open weights other 10.7B parameters transformers [View model](https://savrn.com/models/qwen2-5-omni-7b) Model · Any to any ### [Qwen2.5-Omni-7B-AWQ](https://savrn.com/models/qwen2-5-omni-7b-awq) [Qwen](https://savrn.com/model-publishers/qwen) Qwen2.5-Omni is an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. This model card introduces a series of enhancements designed to improve the Qwen2.5-Omni-7B's operability on devices with constrained GPU memory. Key optimizations include: Implemented 4-bit quantization of the Thinker's weights using AWQ, effectively reducing GPU VRAM usage. Enhanced the inference pipeline to load model weights on-demand for each module and offload them to CPU memory once inference is complete, preventing peak VRAM usage from becoming excessive. Converted… Open weights other 10.7B parameters transformers [View model](https://savrn.com/models/qwen2-5-omni-7b-awq) Model · Any to any ### [gemma-4-E4B-it](https://savrn.com/models/gemma-4-e4b-it) [Google](https://savrn.com/model-publishers/google) Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from… Open weights apache-2.0 8B parameters 131,072 tokens transformers [View model](https://savrn.com/models/gemma-4-e4b-it) ## OpenBMB [All models and datasets](https://savrn.com/model-publishers/openbmb) ## Versions - [503e754207c9](https://savrn.com/models/minicpm-o-4-5/versions/503e754207c9) · current 2026-09-18 ## Explore More - [All any to any models](https://savrn.com/models/tasks/any-to-any) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-19. - [Hugging Face record](https://huggingface.co/openbmb/MiniCPM-o-4_5) - [How the hub is built](https://savrn.com/model-hub/methodology)