DeepSeek v4.1 Flash is just another level 🤯 huggingface.co/deepseek-ai/... - Asymmetric Causal-Encoder-Decoder: 550B MoE, input 8B / output 16B - Native vision merged into one endpoint - KV cache crushed: ~1/4 the HBM vs last one, 437× smaller than their first model
Who's Who of AI
adinayakup.bsky.social
Why they matter
Researcher with public evidence across AI research.
- AI signals
- 16 past 30d
- Sources
- 1 distinct domains
- Discussions
- 0 past 30d
- Latest signal
- 3d ago
AI Research @Hugging Face 🤗
Contributing to the Chinese ML community.
What they're sharing
deepseek-ai/DeepSeek-V4.1-Flash · Hugging Face huggingface.co
baidu/Unlimited-OCR · Hugging Face huggingface.co
Paper page - Orca: The World is in Your Mind huggingface.co
GLM-5.2: Built for Long-Horizon Tasks huggingface.co
GLM-5.2 - a zai-org Collection huggingface.co
MiniMaxAI/MiniMax-M3 · Hugging Face huggingface.co
PP-OCRv6 - a PaddlePaddle Collection huggingface.co
baidu/Unlimited-OCR-2 · Hugging Face huggingface.co
Spark-X2.6 - a XHToken Collection huggingface.co
MiniCPM-V 4.7 - a openbmb Collection huggingface.co
deep-principle/Hydrogen-1.0 · Hugging Face huggingface.co
moonshotai/Kimi-K3.1-Pro · Hugging Face huggingface.co
Articles & links
Unlimited-OCR 🔥New OCR from Baidu huggingface.co/baidu/Unlimi... It can parse hundreds of pages in a single pass while maintaining stable speed. The key is R-SWA (Reference Sliding Window Attention), which keeps KV cache constant during decoding. 🏆 93% on OmniDocBench 📈 +6% ov…
BAAI just released the Orca paper 🔥 ( weights coming soon ) huggingface.co/papers/2606.... A Multimodal Latent World Model: it learns the world itself first, and text/images/actions are just different ways to read it out 💡
AI Weekly's analysis
→
- Orca pretrains on 125K hours of video and 160M event annotations using a single Next-State-Prediction objective on a frozen Qwen3.5 backbone.
- Orca-4B averages 51.8 across MVBench, TemporalBench, 3DSRBench, and SWITCH, ahead of Qwen3.5-4B's 46.7 at the same size.
- On a real-robot out-of-distribution test Orca reports 36.6% versus π₀.5's 27.6%, despite using no action labels in pre-training.
Read full analysis →
Really cool to see the GLM 5.2 blog on Hugging Face 🔥 huggingface.co/blog/zai-org...
AI Weekly's analysis
→
- GLM-5.2 scores 74.4% on FrontierSWE, within one point of Claude Opus 4.8's 75.1%, released under MIT license.
- The 753B model uses an IndexShare architecture that cuts per-token compute by 2.9x at 1M token context length.
- Z.AI claims GLM-5.2 is the highest-ranked open-source model across all three long-horizon coding benchmarks tested.
Read full analysis →
GLM 5.2 is here 🔥 huggingface.co/collections/... ✨ 753B - 1M context ✨ MIT license ✨ GLM IndexShare: reuses the indexer across layers, 2.9x fewer FLOPs/token at 1M ✨ AIME 2026: 99.2 (beats GPT-5.5, Claude Opus 4.8) ✨ vLLM / SGLang / Transformers supported
AI Weekly's analysis
→
- Z.ai released GLM-5.2 on Hugging Face, a 753B-parameter open weights model under an MIT license with a 1M-token context window.
- An IndexShare design reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9x at 1M context length.
- Self-reported scores include 62.1 on SWE-bench Pro, 82.7 on Terminal Bench 2.1, 99.2 on AIME 2026, and 91.2 on GPQA-Diamond.
Read full analysis →
MiniMax-M3 just dropped 🔥 huggingface.co/MiniMaxAI/Mi... ✨ 428B / 23B active ✨ 1M context ✨ MiniMax Sparse Attention (MSA) And it’s not just weights! - paper: huggingface.co/papers/2606.... - kernel: huggingface.co/kernels/Mini... - Transformers support Love how this was relea…
PP-OCRv6 just released by Baidu huggingface.co/collections/... ✨ tiny 1.5M / small 7.7M / medium 34.5M ✨ 48+ languages ✨ Supports handwritten/printed/industrial/screen and card text ✨ Edge friendly deployment
Unlimited OCR 2 🔥 New 3B OCR from Baidu huggingface.co/baidu/Unlimi... - MIT license - Unlimited horizon document parsing - SR SWA attention with nearly constant per step memory & latency - SOTA on 100 page MarathonDocBench
After its 4B model, SparkLLM is back again with an even smaller one: Spark-X2.6-1.7B huggingface.co/collections/... - 3 thinking modes: none, low & high - Up to 1M context - Apache 2.0 + GGUF/int4/int8/FP8 - Leading decoding throughput vs similarly sized models in reported tests
✨ Both support: - Image, multi-image & video understanding - Efficient 4×/16× visual token compression - iOS, Android & HarmonyOS - Apache 2.0 huggingface.co/collections/...
AI Weekly's analysis
→
- The 35B-A3B variant is the first MoE in the MiniCPM-V series, claiming 13+ tokens/s with 3 GB of memory on an iPhone 17 Pro Max.
- OpenBMB says the 35B Instruct variant scores higher than Qwen3.6 35B-A3B at lower inference time; the 1B substantially outperforms Qwen3.5 0.8B.
- Release is Apache 2.0 with FP8, AWQ, GPTQ and GGUF quants, and day-0 support for Rockchip, Houmo, Intel, AMD, and Arm chips.
Read full analysis →
Hydrogen 1.0 🔬 Impressive open model built for scientific discovery huggingface.co/deep-princip... - Based on GLM 5.3 - 1M context window - MIT license + with BF16, NVFP4,MXFP4 variants
Kimi K3.1 Pro 🔥 huggingface.co/moonshotai/K...
Their network
In adinayakup.bsky.social's orbit
Center = adinayakup.bsky.social. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.
Are you adinayakup.bsky.social? Show it.
Add the Who’s Who of AI badge to your site or bio. It links back to this profile.
Markdown: [](https://aiweekly.co/whos-who/person/adinayakup-bsky-social)