#MiniCPM5-2B
LightNews LightNews
adinayakup.bsky.social
When everyone is releasing big models, OpenBMB keeps shipping small but strong one.

Here is the latest one: MiniCPM5-2B 🔥
huggingface.co/openbmb/Mini...

- Dense 2B
- Full data pipeline open as well: web, code, agent, RL
- 131 context
- Apache 2.0
September 7, 2026 at 1:58 PM
og-programmer.bsky.social
I been using Spark-X2.5:4b and miniCpm5:2b, they do a decent job
September 27, 2026 at 2:00 PM
gigazine.net
小型かつ高性能で日本語応答も可能な中国製AIモデル「MiniCPM5-2B」が登場、2BモデルなのにGemma 4 12Bと同等のベンチマークスコア
https://gigazine.net/news/20260908-openbmb-minicpm5-2b/
小型かつ高性能で日本語応答も可能な中国製AIモデル「MiniCPM5-2B」が登場、2BモデルなのにGemma 4 12Bと同等のベンチマークスコア
中国のAI企業であるOpenBMBがAIモデル「MiniCPM5-2B」を2026年9月7日に公開しました。MiniCPM5-2Bは20億パラメーター規模のモデルですが、40億パラメーター規模のモデルより高い性能を備えています。
gigazine.net
September 8, 2026 at 1:48 AM
spacesofplay.com
Local LLM playing breakout. It's quite bad on this setup, but fast! (Not sure text representation is ever good here.)

The state is given as "The ball is to the left of the paddle, drifting away from the paddle." and the model outputs {left, stay, right} state.

(MiniCPM5-2B-Q4_K_M, M1 GPU)
September 21, 2026 at 10:34 AM
spacesofplay.com
Finally found an Atari game that LLMs can play, here's MiniCPM5-2B at Q4_K_M playing Hangman: (all local, all minc code, except for atari env which is in python).
September 25, 2026 at 8:25 AM
jauntywunderkind.bsky.social
i'd love to see this compared to MiniCPM5-2B.
bsky.app/profile/adin...

i have no idea how fast it runs. one of the nice things about LFM models is they are usually speedy+efficient. lfm2.5-2.6B is free on openrouter (!) via liquid themselves, >110 tps.
adinayakup.bsky.social Adina Yakup @adinayakup.bsky.social · 20d
When everyone is releasing big models, OpenBMB keeps shipping small but strong one.

Here is the latest one: MiniCPM5-2B 🔥
huggingface.co/openbmb/Mini...

- Dense 2B
- Full data pipeline open as well: web, code, agent, RL
- 131 context
- Apache 2.0
September 15, 2026 at 9:12 PM
techlatest.bsky.social
The interesting part: it runs locally on an Ubuntu VM.

Read the full walkthrough here: medium.com/@techlatest....
From Raw Logs to Root-Cause Clues with OpenClaw & MiniCPM5–2B
AI tools, developer platforms, and infrastructure generate an enormous amount of information every day. Application logs are a perfect…
medium.com
September 18, 2026 at 6:10 PM
courionai.bsky.social
MiniCPM5-2B is 2.5 billion parameters, Apache 2.0, holds 131,000 tokens of context and runs in Ollama or LM Studio on the laptop you already own.

courionai.com/news/2026-09-09-minicpm5-2b-on-device-model

#AI #LocalAI #OpenWeights
A small handheld device sending out wide confident rings of energy in the foreground while a lone server cabinet sits tiny on the horizon
September 22, 2026 at 11:23 PM
jauntywunderkind.bsky.social
this was supposed to be a MiniCPM5-2B stan piece!
bsky.app/profile/adin...
adinayakup.bsky.social Adina Yakup @adinayakup.bsky.social · 20d
When everyone is releasing big models, OpenBMB keeps shipping small but strong one.

Here is the latest one: MiniCPM5-2B 🔥
huggingface.co/openbmb/Mini...

- Dense 2B
- Full data pipeline open as well: web, code, agent, RL
- 131 context
- Apache 2.0
September 15, 2026 at 9:07 PM
techlatest.bsky.social
What happens when a small local AI model gets a real debugging job?

We built a local log-analysis workflow with OpenClaw + Ollama + MiniCPM5-2B that turns raw application logs into errors, timelines, evidence, possible causes, and troubleshooting steps.

#OpenClaw #Ollama #AI #DevTools
September 18, 2026 at 6:10 PM
aigamematome-eng.bsky.social
[Shocking] Evolution of Music Generation AI and Local Language Models! Latest Trends Defying Generative AI Common Sense

▼Read More
https://gaming-matome.com/archives/50262
[Shocking] Evolution of Music Generation AI and Local Language Models! Latest Trends Defying Generative AI Common Sense - AIゲームまとめ
This Generative AI Weekly introduces notable latest technologies and research. 'Udio E2', a music generation AI comparable to Suno v5 and v6, has emerged, allowing sheet music creation from lyrics and style prompts, direct human editing, and cover generation. OpenBMB released 'MiniCPM5-2B', a 2.1B parameter local language model running on smartphones and PCs with high performance in programming, math, and long-text comprehension. DeepSeek announced 'DeepSeek-V4-Flash', supporting up to 1 million tokens of context and outperforming Claude Opus 5 on some agent benchmarks. Sony AI and other research teams announced 'HacKen', an AI system that predicts scientific discoveries from vast historical papers, successfully discovering and evidencing unknown genetic interactions in biomedicine.
gaming-matome.com
September 20, 2026 at 6:52 AM
ryutok.bsky.social
llama.cppを最新を追っかけるのからreleaseで動かすのに運用を変えて、RaspberryPi5にMiniCPM5-2B-q8_0とLFM2.5-2.6B-QAD-Q4_0を入れて、母艦にはOrnith-1.5-9B-Q8_0を入れて、llama.cppの-ctk,-ctvをq4_0からiq4_nlに変更して、、、

なんとなく、賢くなった気がするｗ（おっさんの感想
September 20, 2026 at 4:43 AM
clawbeat.bsky.social
Press enter or click to view image in full size AI tools, developer platforms, and infrastructure generate an enormous amount of information every day. Application logs are a perfect example. When an…
From Raw Logs to Root-Cause Clues with OpenClaw & MiniCPM5–2B
Press enter or click to view image in full size AI tools, developer platforms, and infrastructure generate an enormous amount of information every day. Application logs are a perfect example. When an…
clawbeat.co
September 18, 2026 at 3:35 PM
oludai.bsky.social
Open-weight 2B models rarely see 102k monthly downloads. openbmb/MiniCPM5-2B just did it with an Apache-2.0 license 🌟

https://olud.ai/#leaderboard

#OpenSource #AI #HuggingFace
September 13, 2026 at 8:00 PM
huggingfacetrends.bsky.social
今日のHuggingFaceトレンド

openbmb/MiniCPM5-2B
MiniCPM5-2Bは、デバイス上やローカルなどのリソース制限がある環境での利用を目的とした2BクラスのTransformerモデルです。
同規模のオープンソースモデルの中で最高水準の性能を達成しており、コーディング、数学、長文理解、ツール利用、エージェントタスクなどに強みを持ちます。
openbmb/MiniCPM5-2B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
September 14, 2026 at 4:14 PM
future-shock.ai
OpenBMB released MiniCPM5-2B under Apache 2.0, along with base, mid-training and instruction-tuned checkpoints plus several training datasets. Local builders can choose where to start instead of adapting only the final model.

https://huggingface.co/openbmb/MiniCPM5-2B
openbmb/MiniCPM5-2B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
September 8, 2026 at 5:35 PM
parisien.cc
Im using MiniCPM5-2B rn and im really fucking impressed

It’s a tiny model and yet it’s able to do a deep research iteratively AND taking notes at the same time in an external database.

It’s currently at its 20th tool call moving onto phase 2 of its deep research.
September 17, 2026 at 11:19 PM
huggingfacetrends.bsky.social
今日のHuggingFaceトレンド

openbmb/MiniCPM5-2B
リソース制限のある環境やオンデバイス向けに設計された20億パラメータの言語モデルMiniCPM5-2Bを公開するためのリポジトリです。
同クラスのオープンソースモデルにおいて最高水準の性能を実現し、特にコーディングや数学、長文理解などのタスクで高い能力を発揮します。
openbmb/MiniCPM5-2B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
September 10, 2026 at 2:09 PM
atbart.bsky.social
53.9 average across 34 benchmarks from a 2B edge model. MiniCPM5-2B tops open-source sub-4B on the ArtificialAnlys index. The agentic score running on-device is the part worth watching. small language model edge inference open source rl

da.gd/YNsLwz
September 9, 2026 at 8:27 AM
oludai.bsky.social
🤗 Climbing on Hugging Face right now: openbmb/MiniCPM5-2B

⬇ 67.6k downloads / 30 days · ♡ 1.2k · text generation · Apache-2.0

Open-weight models ranked by real-world usage, every day:
https://olud.ai/#leaderboard

#OpenSource #AI #HuggingFace
September 12, 2026 at 9:00 PM
trendai.bsky.social
Master local AI deployment! Learn how to run Spark-X2.5-4B and MiniCPM5-2B using Gemma family inference engines. High performance, even with minimal RAM/VRAM requirements. #AI #LLM #LocalAI #MachineLearning #Deployment https://ai.dappcrypto.org/r/eE
Master local AI deployment! Learn how to run Spark-X2.5-4B and MiniCPM5-2B using Gemma family inference engines. High performance, even with minimal RAM/VRAM requirements. #AI #LLM #LocalAI #MachineLearning #Deployment https://ai.dappcrypto.org/r/eE
September 8, 2026 at 7:33 PM
tigercapitalresear.bsky.social
substack.com/@princetonai...
Tiger VC Newsletter (@princetonainewsletter)
🤖They released MiniCPM5-1B: a dense 1-billion-parameter model that doesn’t just compete with 2B-class models, it beats them. On the Artificial Analysis Intelligence Index (the most respected independ...
substack.com
May 27, 2026 at 12:20 AM
oludai.bsky.social
🤗 Climbing on Hugging Face right now: openbmb/MiniCPM5-2B

⬇ 2.9k downloads / 30 days · ♡ 998 · text generation · Apache-2.0

Open-weight models ranked by real-world usage, every day:
https://olud.ai/#leaderboard

#OpenSource #AI #HuggingFace
September 10, 2026 at 9:00 AM
oludai.bsky.social
MiniCPM5-2B from openbmb is climbing Hugging Face with 2.9k monthly downloads—an Apache-2.0 text gen model small enough to run anywhere, yet pulling real attention. See its daily usage rank now:

https://olud.ai/#leaderboard

#OpenSource #AI #HuggingFace
September 10, 2026 at 10:08 PM
barredspirals.comint.su
nono qfn is still in reasoning mode, and there I think it's helping compensate for the quantization a bit. the model i'm running in instruct mode is minicpm5-2b (at q8). there's also no art yet (we're still working on the daemon) but per previous tests it's pretty good at 3d modelling at least
September 14, 2026 at 7:55 AM