DeepSeek V4’s 1M-Token Gauntlet: Can China Out-Context GPT-4o?
Audio in Mandarin Chinese · English transcript below
⚡ DeepSeek slashes 1M tokens to cents, Vision Banana crowns segmentation, Isomorphic’s AI drug hits humans, Cohere merges across oceans—retirees rave with ChatGPT as AI accelerates!
Today's Top 3 Headlines
- AI Industry News
DeepSeek Drops 1M-Token V4 LLM, Taking Aim at U.S. Rivals
**DeepSeek V4** debuts with **1M-token context**, topping U.S. rivals in code, inference and long-memory while costing **<50%**; analysts call the LLM a leap for complex, doc-heavy Agents.
Source ↗ - AI Industry News
DeepSeek’s Dirt-Cheap LLM Just Shook Silicon Valley—Gizmodo
DeepSeek drops V4 Pro at **<1/5** of Gemini 3.1 Pro’s price, betting cheap compute can lure devs and corporates, staking a seat in the global LLM race.
Source ↗ - AI Industry News
DeepSeek V4 First Benchmarks: Breakthrough or Bust? (Full Leaderboard Inside)
**DeepSeek V4** open-source preview drops: **16T total/49B active** Pro & **284B/13B** Flash, **1M-token** context, MIT license. Arena ranks Pro #3 open, #14 overall; Vals crowns it #1 open, **~10×** leap over V3.2, **0.07%** behind Kimi K2.6. Pricing: Flash **$0.28**, Pro **$3.48** per 1M out—**>99%** cheaper than Claude Opus. Team warns 3-6 mo gap to closed rivals, limited compute.
Source ↗
+7 more headlines
- DeepSeek Unveils Major AI Model Upgrade
- Cohere Merges With German Startup to Forge Transatlantic AI Powerhouse
- DeepMind Spin-Off’s AI-Designed Drug Cleared for First-in-Human Trials
- Retirees vs. AI: How the Reddy Family Uses ChatGPT to Reinvent Life After Work
- I Killed Claude: Token Burn, Quality Drop, Zero Support
- Op-Ed: AI Doomers Must Face the Hard Truth About Water Policy
- Google DeepMind Drops Vision Banana: Instruction-Tuned Image Generator Beats SAM 3 at Segmentation, Depth Anything V3 at Metric Depth