DeepSeek Launches New AI Model as Funding Rumors Spread
Listen to the full version
Chinese artificial intelligence (AI) startup DeepSeek launched its highly anticipated V4 open-source foundational model on Friday, upping the ante in a race with global peers to develop a technology that underpins data centers and AI applications.
The DeepSeek-V4 model comes in Pro and Flash versions. The Pro version features 1.6 trillion total parameters with 49 billion active parameters, making it an open-weights model trained with the most parameters. The Flash version contains 284 billion total parameters with 13 billion active parameters to provide more cost-effective API services. Both versions support a context window of up to one million tokens.
Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.
Save an extra $50. Introductory offer for new readers. Subscribe now.
- RELATED
- DIGEST HUB
Digest Hub
Back
- DeepSeek launched V4 open-source model: Pro (1.6T total/49B active params), Flash (284B/13B), both with 1M token context.
- V4 scores 63.87% accuracy in financial/legal/coding tests, trails Claude Opus 4.6/Gemini 3.1 Pro but matches top models in math/STEM/coding.
- Features Sparse Attention reducing compute 90% (Flash)/70% (Pro) vs V3.2; rumors of >$20B Tencent/Alibaba investment.
- DeepSeek
- DeepSeek, a Chinese AI startup, launched its V4 open-source model with Pro (1.6T total/49B active params) and Flash (284B total/13B active) versions, both supporting 1M-token context. It optimizes compute efficiency (up to 90% reduction) and agent tasks but trails top models like Claude Opus in some benchmarks. Rumors suggest $20B+ valuation talks with Tencent and Alibaba.
- Moonshot AI
- Moonshot AI is a rival to DeepSeek. Its Kimi K2.6 model outperforms DeepSeek-V4 in output speed and average accuracy (V4 at 63.87%) across financial, legal, and coding tests, surpassing rivals like Claude Opus 4.6.
- Zhipu AI
- Zhipu AI is a rival to DeepSeek, with similar models that DeepSeek-V4 did not significantly outperform, according to developers. (23 words)
- VALS AI
- VALS AI is a third-party AI model performance evaluation firm. It reported DeepSeek-V4's average accuracy of 63.87% across financial, legal, and coding tests, lagging behind Claude Opus 4.6, Gemini 3.1 Pro Preview, GPT-5.4, and Kimi K2.6. V4's output speed trails Kimi K2.6.
- Tencent Holdings Ltd.
- Recent rumors suggest Tencent Holdings Ltd. is in talks with Alibaba to invest in DeepSeek at a valuation exceeding $20 billion, though amounts and terms may adjust.
- Alibaba Group Holding Ltd.
- Rumors suggest Alibaba Group Holding Ltd. and Tencent Holdings Ltd. are in talks to invest in DeepSeek at a valuation exceeding $20 billion, though details may adjust.
- MOST POPULAR
Sign in with Google
Sign in with Facebook
Sign in with 财新