← The Polylog AI Intelligence Brief
Morning Edition · Saturday, June 27, 2026
Ai2's OLMo Hybrid Argues Attention Plus Linear RNNs Beats Either Alone
The Allen Institute's 7-billion-parameter hybrid interleaves transformer attention with Gated DeltaNet, reaching OLMo 3's accuracy on the Massive Multitask Language Understanding (MMLU) benchmark using 49 percent fewer tokens.
The Allen Institute for AI published the work behind OLMo Hybrid, a 7-billion-parameter model that interleaves standard transformer attention layers with Gated DeltaNet, a modern linear recurrent neural network. The Russian-language summary…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
More from this edition
- OpenAI Previews GPT-5.6 Sol Under Government-Gated Release to 20 Partners
- DeepSeek Open-Sources DSpark, Pushing V4 Inference Speed Up Sharply
- U.S. Eases Anthropic Block, Allowing Mythos 5 for Select Cyber Defenders
- EU Picks Italy's Domyn to Build a 400-Billion-Parameter Sovereign Model
- The Frontier Throughput Race Moves to Custom Inference Silicon
- Bank of Korea Finds AI Saves 1.5 Hours a Week, Lifts Output Near Zero
- Meta Updates SAM 3 to 3.1, Targeting Faster Real-Time Video Tracking
- Anthropic Pushes Policy Overhaul for an 'AI Exponential'
- Anthropic Ships Claude Tag for Team Workflows
- Meta Previews Muse Spark From Its Superintelligence Labs
- Meta Details How It Builds and Tests Advanced AI at Scale