Estética IA · tendencias visuales · arte generativo Prompts • herramientas • inspiración • aesthetic
Joined August 2021
- Tweets5.7K
- Following304
- Followers10.6K
- Likes14.3K
✦ VISUAL AI ✦ retweeted
An AI that can read a crumpled invoice or a tilted photo as if it were a clean PDF...
MiniCPM-OCR 1.0 is open source and delivers 3.62× the throughput of MinerU2.5 Pro.
How many hours do you spend manually copying information from documents?
ALT Throughput chart: MiniCPM-OCR 1.0 processes 1.217 pages per second, ahead of 13 other OCR models tested on an RTX 5090
Your OCR says it’s done—but did it get everything right? 🤔 In high-stakes workflows, even a small error can be costly.
🚀 Meet the MiniCPM-OCR suite: an open-source solution to parse fast and guard the results.
Two new models:
🔹 MiniCPM-OCR 1.0 — SOTA OCR performance and speed at 1B scale. Compact, accurate, and fast.
🔹 MiniCPM-OCR Jev — A 2B decision model that helps assess parsing accuracy, with performance approaching that of Qwen3.5-397B-A17B on JevOCRBench.
⚡ What’s new?
Small model, SOTA results: MiniCPM-OCR 1.0 scores 93.68 on Real5-OmniDocBench and 54.7 on OCRBench v2.
A sharp goalkeeper 🛡️: MiniCPM-OCR Jev is model-agnostic, checking text, formulas, and tables from different OCR parsers. It achieves 93.6 on JevOCRBench, outperforming JEV-35B-A3B (90.1) and approaching Qwen3.5-397B-A17B (93.8).
Up to 128 pages per minute 🤯: With DSpark, MiniCPM-OCR runs at 1.49× the speed of PaddleOCR-VL-1.6 and 3.62× that of MinerU2.5 Pro. The SDK’s concurrency optimizations further boost throughput to as much as 128 pages per minute.
Deploy where you work: Day-0 support for Arm, Houmo AI, Intel, and NVIDIA, plus 10 more chip vendors through FlagOS.
Ready for your stack: vLLM, SGLang, llama.cpp, and Ollama, with quantized models, fine-tuning support, and Agent Skills.
🔗 GitHub: github.com/OpenBMB/MiniCPM-O…
🤗 Hugging Face:
MiniCPM-OCR 1.0：huggingface.co/openbmb/MiniC…
MiniCPM-OCR Jev：huggingface.co/openbmb/MiniC…
👉 Try the Demo：platform.modelbest.ai/consol…
Use the API：platform.modelbest.ai/consol…
este es el tiipo de lanzamiento open source que ahorra miles de horas a las empresas sin hacer ruido
✦ VISUAL AI ✦ retweeted
POV you're at the beach and you open your laptop because Cursor needs a yes to install a package 🏖️
The agents from tryAlanAI keep going in the cloud while you're away. My tan thanks them in advance.
🤝 Paid partnership
Today we're launching @tryAlanAI so developers can stop babysitting their coding agents.
We kept running into the same problem ourselves.
Even as agents got better at writing code, we were becoming glorified button-pushers for our own AI: starting every run, feeding them context, unblocking failures, and checking their work.
We built Alan to change that.
Mi agente de código tiene más notificaciones pendientes que el grupo de WhatsApp de mi familia 😂
@tryAlanAI, os necesito.
🤝 Paid partnership
Today we're launching @tryAlanAI so developers can stop babysitting their coding agents.
We kept running into the same problem ourselves.
Even as agents got better at writing code, we were becoming glorified button-pushers for our own AI: starting every run, feeding them context, unblocking failures, and checking their work.
We built Alan to change that.
✦ VISUAL AI ✦ retweeted
Me alegra que esto también tenga en cuenta la experiencia. Conseguirlo al cuarto intento se siente muy diferente a lograrlo al primero.
Today, @JudgmentLabs opens for general access.
We believe long running agent systems will generate more exhaust data than the world can currently process.
Hidden within agent traces are rich records of how to make these systems better: what users ask for, when agents stumble, and how agents interact with the external world.
Judgment is the most performant monitoring and improvement layer for long-horizon agents.
We’re excited to see what you build!
✦ VISUAL AI ✦ retweeted
POV el de ventas te manda "presentacion_FINAL_final_v7_BUENA.pptx" con Comic Sans y el logo de hace 5 años 😭
Ojalá los Templafy Document Agents hubieran existido antes de ese archivo.
AI that knows your company’s content and branding? 👀
With Templafy Team, build and share AI Agents that create editable PowerPoints using your company's knowledge and templates.
Get started 👉 templafy.com/templafy-teams-…
Es una locura.. Este tipo conectó su iPhone al Mac y Claude Code encontró miles de fotos que había borrado hace años, guardadas en una caché oculta.
Capturas de apps bancarias
fotos borradas al instante
y cosas que eliminó dos veces.
Revisa el tuyo ya.
✦ VISUAL AI ✦ retweeted
Every new hire gets a 90-day plan. This week I wrote one for the AI employee on our search team.
I'm Evan, an AI researcher in Boston and NYC working on search, RAG and knowledge systems.
Two lines on the page are already ticked, both from work Viktor did before the plan existed:
→ He read every "not found" from real users in September. 187 of 4,316 queries. Only 14 were actually missing from the docs.
→ He checked our citations. 19 of 154 answers cited a passage that didn't back the claim.
Those two checks used to take me about 4 hours every week. Now I spend 20 minutes reviewing what he flags.
None of those failure types were in our eval set. That's what the plan is for.
Days 1 to 30: learn our search. Every folder, and who owns it.
Days 31 to 60: own the checks. Failed queries read every Monday, a citation sample every week, an owner proposed for the 52 docs nobody owns.
Days 61 to 90: run search quality. Real failures go into the eval set, and every re-index plan waits for my yes.
By day 90: every way our users fail is in our eval set.
House rule: he proposes, I approve. He changes nothing in the index.
An eval set you wrote yourself only asks the questions you already know how to answer.
Hire @viktor_com for your team. Start free.
#AIemployee #RAG #AIsearch
Paid Partnership
🤝 Paid partnership
✦ VISUAL AI ✦ retweeted
llevo meses montando mis reels a trozos en 3 apps distintas y ahora descubro que @usecardboard hace todo el montaje directamente dentro de claude y chatgpt con un plugin..
🤝 Paid partnership
Introducing Cardboard for ChatGPT and Claude.
You never know when your best ideas will find you.
So we made sure that @usecardboard is there, whenever they do.
At your desk, or on the go, wherever you are :)
✦ VISUAL AI ✦ retweeted
Our employee of the month saved us from the vendor we were about to pick.
I lead product on AI tools for creators. A few weeks ago we had to choose a speech-to-text provider, and vendor calls usually go the same way: a few clips, a demo, a gut call, and a sales team quoting their own benchmark.
A benchmark is someone else's audio. Our creators don't sound like a benchmark.
So the bake-off went to Viktor, our AI employee:
→ 97 real clips from our own creators
→ all 4 vendors on the shortlist
→ every clip scored on accuracy, speed and cost per audio hour
→ one sheet, ready before the first sales call
The vendor we were leaning towards came last on our noisiest clips.
He emailed no vendor and signed nothing. He proposes, I decide.
The award isn't for the 97 clips. The award is for the call we didn't get wrong.
Hire @viktor_com for your team. Start free.
#AIemployee #ProductManagement
Paid Partnership
🤝 Paid partnership
