Magic claims its offering matches DeepSeek V4 Pro’s base model at a cost of $500,000, cutting pre-training computing power consumption by 50 times.
55 minutes ago
Beating AI Express (Insight) – AI coding firm Magic has released a pre-training study, claiming a new training scheme achieves pre-training performance nearly on par with DeepSeek V4 Pro Base using just ~$500,000 worth of GB200 compute, with compute requirements roughly 1/50th of the latter. "Base" refers to a foundation model that has only completed pre-training, not undergone post-training steps like reinforcement learning. Magic tested models on unseen code, math problems, and research papers to evaluate their accuracy in predicting subsequent content. The comparison here is limited to base model pre-training performance, and does not mean Magic’s models have matched the full DeepSeek V4 Pro’s chat, coding, or agent capabilities. Magic then scaled the training size 10x, requiring ~$4 million in GB200 compute. This scaled version outperformed DeepSeek, Kimi, and NVIDIA’s latest publicly available base models in the same self-administered test. Magic estimates that reaching the same performance level at DeepSeek V4 Pro’s training efficiency would cost over $100 million in compute. The firm did not disclose the full training recipe, only noting efficiency gains stem from dozens of adjustments to model architecture, optimizer, training objectives, data processing, and other areas. As model weights have not been released, the "50x efficiency" claim cannot be independently verified by external parties.
Three months after stepping down as CEO of DingTalk, reports say Chen Hang, who goes by the nickname Wu Zhao, has joined Qianwen Office.
2 minutes ago
Bitcoin wallet Alby confirms vulnerabilities in its older versions, urges users to update immediately.
2 minutes ago
Bonk Guy: MARSCOIN may be deliberately price-suppressed by market makers to attract short sellers, and could subsequently stage a sudden sharp pump.
2 minutes ago
An entity dormant for six months has built a large position of 13,290 ETH and 6,601 ZEC.
2 minutes ago
The German government plans to impose a 25% capital gains tax on cryptocurrency gains.
2 minutes ago
Founder of Pons: The tax rate for tokens issued on the platform cannot be adjusted after issuance, and the abnormal display is due to terminal routing issues.
2 minutes ago
Hot feeds
A trader profits $448K by monitoring #Binance's new listings!
2024.12.13 17:37:29
Last week, funds have flowed into #Bitcoin, #Ethereum, and #Hyperliquid.
2024.12.16 14:48:36
A $PEPE whale that had been dormant for 600 days transferred all 2.1T $PEPE($52M) to a new address.
2024.12.14 10:35:27
When Elon Musk tweeted about Moltbook, the meme coin MOLT experienced a short-term 30% price surge, hitting a new all-time high of $114 million.
2026.01.31 18:37:29
A smart #AI coin trader made $17.6M on $GOAT, $ai16z, $Fartcoin,$arc.
2025.01.05 16:05:18
A sniper earned 2,277 $ETH ($8.3M) trading $SHIRO within 18 hours!
2024.12.03 23:09:08
MoreHot Articles

How did I turn $1,000 into $30,000 with smart money?
2024.12.09

10 promising AI Agent cryptos
2024.12.05

The 30-Year-Old Entrepreneur Behind Virtual, a Multi-Million Dollar AI Agent Society
2025.01.22

10 smart traders specializing in MEMEcoin trading on Solana
2024.12.09

A trader lost $73.9K trading memecoins in just 3 minutes — a lesson for us all!
2024.12.13

What is $SPORE? Let us take you through the on-chain records to show you how it works.
2024.12.25

