Satu juta token dalam konteksnya digunakan untuk meminta 4 GB VRAM per aliran dalam cache KV saja.

DeepSeek-V4 reduces it to 82 MB: a 98 percent collapse in memory footprint. Here is the three-part architectural co-design that made it possible: 1. Compressed Sparse Attention + FP4

Alcides TicllaAug 17

DeepSeek V4 Pro baru saja mencetak skor 87,9 di Terminal Bench 2.1.

That matters because this isn’t another “AI answers questions better” benchmark. It tests whether an AI can actually finish tasks inside a computer terminal. And DeepSeek built V4 Pro around exactly that.

Julian Goldie SEOAug 172

DeepSeek-V4 Pro Max vs Claude Opus 5 vs Grok 4.5 Maks

Testing 2M+ Token Context Retrieval & Complex Code Synthesis: 🔹 DeepSeek-V4 Pro Max: Fastest inference & cost-efficiency for massive codebases (98.4% Needle-In-A-Haystack accuracy). 🔹 Claude 5 Opus: Unrivaled structural

AI Cyber DeskAug 17

DeepSeek Harness adalah memanfaatkan agen di mana setiap komponen merupakan plugin, dibangun di Cordis untuk komposisi spatiotemporal.

Jalankan melalui `npx deepseek-ai/dsh web` untuk mendapatkan UI Web lokal di port 3080. Jelajahi di sini:

GitHub Projects CommunityAug 1771