LEANN shrinks a 60M-chunk RAG index from 201 GB to 6 GB
MarkTechPost: Meet LEANN: An Open-Source Vector Index that Shrinks a 60M-Chunk RAG Index from 201 GB to 6 GB
UC Berkeley’s Sky Computing Lab built LEANN, an open-source vector index that stores almost no embeddings. The LEANN vector index recomputes embeddings at query time instead of saving them to disk. The repo benchmark shows a 60 million chunk Wikipedia index dropping from 201 GB to 6 GB. The research paper won the Best Research Paper award at MLSys 2026. So, it brings semantic search and RAG over personal data, on a laptop. TL;DR Size: 6 GB for 60M chunks (repo); 4 GB in the paper&#
Get smart on it
LEANN is an open-source vector index that dramatically reduces storage requirements for retrieval-augmented generation (RAG) systems by recomputing embeddings at query time rather than storing them. A benchmark showed a 60-million-chunk index shrink from 201 GB to 6 GB while maintaining the same accuracy as traditional methods. The trade-off is slower search speed: LEANN retrieval takes 2.48 seconds compared to 0.05 seconds for the standard HNSW approach, though this extends overall pipeline time by only a few seconds since generation typically takes around 20 seconds. This makes semantic search and RAG feasible on laptops and personal computers.
Explainer: How to Actually Read an AI Benchmark (Without Being Fooled)
Relevant courses
Want to go deeper than the news? Explore live, cohort-based AI courses taught by practitioners.
Browse AI courses on MavenRelated news
Introducing GPT-6.1 Terra
Bringing fast and capable computer use to everyone
Also reported by SiliconANGLE, MarkTechPost and The Decoder
Genie 4: A new generation of world simulation
Genie 4: our new world model unlocking unprecedented realism in real-time simulation.
Also reported by TechCrunch, Ars Technica and SiliconANGLE
Google’s next Genie AI model generates more realistic experiences
Google announced today its Genie 4 world model for generating interactive experiences using AI. The company says the model has four main updates: It allows for more realistic worlds, lets users update an experience in real time with a natural language prompt, can generate audio, and uses Gemini 3.8 Flash under the hood to help with its simulations. It also can create experiences that are about two minutes long, up from one minute for the previous Genie 3 model. Google is pitchi
Moonshot AI Releases Kimi K3.1 Pro: A 2.8T Total/104B Active Params (MoE) Model with 1M Context
Moonshot AI has just released Kimi K3.1 Pro, the first model in its K3.1 series. Kimi K3.1 Pro is a 3T-class open-weight* model with a 1-million-token context window. Moonshot team positions it against the strongest proprietary models on coding, agentic and knowledge benchmarks. For developers, it is cheaper API pricing than K3 and open weights yet not available. *open weights are expected to release in November 2026. TL;DR Size: 2.8T total, 104B active parameters (MoE), 1M-token context
Also reported by SiliconANGLE and The Decoder
Build specialized multimodal models with LFM3
Organizations can now create custom AI systems combining text, images, and video without building from scratch, lowering barriers to deployment.
Also reported by MarkTechPost
Mirendil Launches Fern 0.1, Its First Model, With 10M-Token Context
Mirendil has released Fern, its first AI model. Mirendil Fern is a from-scratch coding model with a context window of up to 10M tokens. Mirendil explains how that equals about 23 MB of code, or roughly half a million lines. That is a codebase on the scale of Megatron-LM, held in context at once. Fern 0.1 is the first version, built from Mirendil-Base after a short round of post-training. For developers, its like, you load a whole repo, a full training log or days of agent traces at once. TL;