China's Big Fund Leads DeepSeek's First External Round at $45 Billion Valuation
In court, OpenAI president Greg Brockman revealed that Elon Musk demanded 51% control of the startup to fund a Mars city — and stormed out when they refused.
DeepSeek Chat
Most RecentDeepSeek's whale opens its eyes in tease of new vision model
A DeepSeek researcher posted a cryptic teaser with a whale mascot that goes from blindfolded to seeing, hinting at an upcoming multimodal model.
Kimi's next model could pack 2.5 trillion parameters and 1M context
Leaked specs suggest Kimi's upcoming K3 model will have over 2.5 trillion parameters and a 1M token context window, but the bottleneck isn't technology—it's compute.
Muon optimizer tricks models into overconfidence — and DeepSeek V4 uses it
A new paper finds Muon, an optimizer used by DeepSeek V4, makes models perfectly calibrated on training data but overconfident on new samples.
GPT-5.5 is finally competitive — but OpenAI buried a loss to Opus 4.7
SemiAnalysis tests show GPT-5.5 is once again a top coding model, but OpenAI quietly switched benchmarks after Opus 4.7 beat it — and the fine print tells the real story.
DeepSeek just cut its API cache prices to a tenth
DeepSeek has permanently slashed cache hit prices for its V4 API models to one-tenth of the original cost, with V4-Pro getting an even steeper discount through early May. The move is especially impactful for agentic coding workflows where cache hit rates are high.
US State Department Cables Accuse Chinese AI Startups of Stolen Models
The State Department has sent diplomatic cables to embassies worldwide, warning that Chinese AI companies DeepSeek, Moonshot AI (Kimi), and MiniMax are using 'distillation' to steal US AI models. Beijing calls the allegations 'groundless'.
DeepSeek V4-Pro tops GPT-5.4 on coding, but lags in knowledge and long context
DeepSeek's open-source V4-Pro-Max set a record on Codeforces, beating GPT-5.4, but fell short on knowledge and long-context benchmarks against Gemini and Opus.
DeepSeek V4 makes 1M context standard across all models
DeepSeek's new V4 series makes 1M-token context standard across all tiers, with V4-Pro matching top closed-source models in reasoning and openly acknowledging gaps against Anthropic's Opus 4.6.
DeepSeek drops V4 preview
DeepSeek open-sources its V4 preview under MIT license: 1.6 trillion parameters, 1M token context, and memory usage slashed to a tenth of V3.2.