k
keynews.ai
keynews.ai
FrontierThe story, in brief

DeepSeek's new models are so efficient they'll run on a toaster ... by which we mean Huawei's NPUs

DeepSeek V4 cuts inference costs to a fraction of R1 — now optimized for Huawei NPUs.

Reporting by The Register AI/ML 1 min brief
Illustration of a transparent lens revealing connected networks across layers of paper.
Exploring the next frontier of AI research.AI illustration by KeyNews
The KeyNews take

Why it matters

DeepSeek's new efficiency gains signal a shift in the cost-performance equation for model deployment, with implications for who can afford to run frontier AI at scale.

The key facts

4 to know

  1. DeepSeek V4 now in preview

  2. Inference costs reduced significantly vs. R1

  3. Optimized for Huawei NPU hardware

  4. Efficiency improvements enable deployment on resource-constrained hardware

Go to the source

The Register AI/MLtheregister.com

Publisher excerpt: Now available in preview, DeepSeek V4 cuts inference costs to a fraction of R1
Read original report
Back to today's editionMore frontier news

Keep reading

Related stories

More from Frontier
Frontier

Our framework for reporting model misalignment

OpenAI publishes first systematic framework for tracking model misalignment — and reveals six cases of unexpected behavior.

OpenAI BlogFrontier

Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear

Salesforce + Nvidia just shipped a reasoning model that does sales work out of the box. The frontier labs should be nervous.

TechCrunch AIFrontier

Nums AI Releases Causilo: A Tabular Foundation Model That Tops TabArena Among Single Models

Nums AI's Causilo tops TabArena benchmark—beating Google's TabFM and LG's EXAONE in the race for tabular foundation models.

MarkTechPost