← The Polylog AI Intelligence Brief
Morning Edition · Sunday, July 19, 2026Published at 1:30 AM EDT · New York
Li Auto's 35B Mach-Mind-4-Flash Reports Matching 120B-Class Models on Math and Code
The mixture-of-experts model activates just 3 billion parameters and claims 92.7 on AIME 2026 and 80.9 on LiveCodeBench, gains it credits entirely to post-training.
The AI division of Chinese electric-vehicle maker Li Auto published a technical report on Mach-Mind-4-Flash, a mixture-of-experts model with 35 billion total parameters and only 3 billion active at inference. As flagged in Russian-language…
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.
Part of a tracked trend
Frontier Model Efficiency Gains
Capability per unit of training and inference compute keeps improving, letting newer models match prior frontier performance far more cheaply and gradually loosening the link between raw scale and capability.
Townsquare
Start a discussion in Townsquare.
More from this edition
- AI Inference Prices Split Roughly 100-to-1 as OpenAI Pushes a New Value Metric
- White House Moves to Decide Who Gets Frontier Models From OpenAI and Anthropic
- Meta Opens Its First Paid Model API With Muse Spark 1.1
- Meta's Brain2Qwerty v2 Decodes Typed Sentences From Brain Signals at 61% Word Accuracy
- Anthropic and Rivals Propose a Shared Scale for Rating AI Jailbreak Severity
- Meta Ships Muse Image and Previews Muse Video as Agentic, Editable Generators
- ETH Zurich Demonstrates a Quantum Computer With Mechanical Working Memory
- Anthropic Bundles Fable 5 Into Max and Team Premium at Half Quota
Comments
0No comments yet.