Models & Labs
Subquadratic claims breakthrough in LLM efficiency
MIT Technology Review AImedium confidence
Why it matters
- →SubQ's sparse attention approach could drastically reduce computational costs for LLMs.
- →If validated, this model could lead to more efficient AI systems, impacting various industries.
- →The breakthrough challenges the dominance of dense attention models, potentially shifting AI development paradigms.
Subquadratic, an AI startup from Miami, claims to have developed a new large language model, SubQ, that overcomes a longstanding computational bottleneck. The model reportedly uses sparse attention to reduce the number of computations, making it faster and cheaper to run than traditional models. Independent tests by Appen support these claims, showing SubQ's potential to process large amounts of text efficiently. While the model is not yet widely available, its performance could significantly impact the future design of LLMs.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1New Method Enhances LLM Training Efficiency
MIT News AI · February 26, 2026 · Related
- 2TurboQuant Introduces Extreme Compression for AI
Google Research Blog · March 24, 2026 · Related
- 3LLMs Optimize Database Query Execution Plans
Together AI Blog · April 3, 2026 · Background
- 4Together AI's Inference Engine Outperforms Competitors
Together AI Blog · May 19, 2026 · Background
- 5llama.cpp b9271 Release Enhances Efficiency
llama.cpp Releases · May 22, 2026 · Background
- 6JetBrains Releases Mellum2: Efficient 12B MoE Model
Hugging Face Blog · June 1, 2026 · Related
- 7llama.cpp b9455 Release Adds Quantized KV Cache
llama.cpp Releases · June 2, 2026 · Background
- 8Google DeepMind unveils DiffusionGemma for faster text generation
Google DeepMind · June 10, 2026 · Background
What happened next
- 9OpenAI and Broadcom unveil LLM-optimized chip
OpenAI · June 24, 2026 · Related
- 10Startups Innovate Beyond Transformers in LLMs
MIT Technology Review AI · August 10, 2026 · Same story
- 11Recall Bottleneck in LLMs: New Google Research Insights
Google Research Blog · August 12, 2026 · Related
- 12Qwen3.8-27B Model Explored and Optimized
Sam Witteveen · August 18, 2026 · Related
- 13Granite 4.2 LLMs Introduced by IBM and Hugging Face
Hugging Face Blog · August 25, 2026 · Background
More from MIT Technology Review AI
More in Models & Labs
Models & Labsmodels
llama.cpp b12029 adds CUDA 13 and Snapdragon support
This release quietly cements llama.cpp as the universal inference runtime by finally supporting NVIDIA's latest CUDA 13 stack alongside AMD's ROCm 10.0. The inclusion of native builds for Linux arm64 with Snapdragon chips, covering CPU, Adreno GPU, and Hexagon NPU, signals a serious push into mobile edge deployment. While Apple Silicon KleidiAI is disabled in this specific build, the broader hardware coverage means developers no longer need to compile from source for most modern accelerators. It’s less about new features and more about removing friction for anyone running local models on bleeding-edge or diverse hardware.
llama.cpp Releases
© The Verge AIModels & Labsother
Apple Intelligence arrives in China via iOS 27.3
Apple is finally unlocking its AI suite for Chinese users with the iOS 27.3 beta, marking a significant regulatory milestone after months of approval delays. The rollout relies on Alibaba’s Qwen model rather than OpenAI, integrating features like notification summaries and visual intelligence directly into the OS. By sidestepping the new Siri in favor of established text and image tools, Apple is prioritizing immediate utility over the full generative experience. This move effectively brings Apple’s AI capabilities to a massive new market, albeit with localized constraints that distinguish it from the global release.
The Verge AI
© Matt WolfeModels & Labsmodels
Google Releases Gemini 4 Carbon for Cost-Efficient Inference
Google launched Gemini 4 Carbon, a new model variant aimed at providing high performance at a lower cost for inference tasks.
Matt Wolfe