StoryFirst seen
Compression cuts the cost of running agents that read far more than they write, where memory overhead has been a core expense.
Everything it touches
1 names 1 voices 1 source posts
The story’s neighborhood
Explore the graph, or choose a connection below.
Follow the evidence
This brief comes from Tech Twitter’s edited headlines. The posts below are the sources grouped with the story; statements in them belong to their authors.
Agents read more than they write, so DeepSeek shrank what they remember. Read the technical breakdown in this week's issue of The Batch. 📰🗜️ 📉 890 bytes of cache per token, 437x smaller than DeepSeek-V1 ⚡ +25% compute per output token from 4K to 1M input 🏆 Flash beats V4-Pro: 39 vs 36 on AA Index, $0.27 vs $0.67/task #DeepLearningAI #LLMs #AIAgentsRead original on X
DeepSeek ships a model with drastically smaller agent memory
DeepSeek ships a model with drastically smaller agent memory
Shared connections
It reportedly uses about 45% fewer output tokens than a comparable model while keeping similar benchmark accuracy.
The test uses 206 multi-step tasks across fields like chip design, biology, and quant trading, showing how far current models fall short on professional work.
The nearly $16 billion round drew backing from Tencent, CATL, and Chinese state-owned institutions.
The round more than doubled its original target and values the company at about $74 billion before the new cash.