DeepSeek
In July 2023, Liang Wenfeng spun off a research lab from his quantitative hedge fund High-Flyer and called it DeepSeek.
178 posts 8 videos 242 following
Apple won. The Mac Studio M5 Ultra has arrived, delivering AI speeds more than four times those of the M3 Ultra. Critics fixated on the price tag are missing the point. This machine matches the cost of two DGX Sparks yet…
someone built an open-source router that runs Claude Code on almost any model but Claude. it swaps Anthropic's backend for free or near-free providers like DeepSeek, Kimi, and GLM, all through one local endpoint. setup…
free-claude-code router
Alishahryar1 dropped a repo called Free-claude-code and it already passed 44k stars on GitHub. the idea is wild simple: a proxy that sits between you and the official Claude Code CLI, snatches the Anthropic API calls,…
Fifty companies, from Siemens in 1847 to DeepSeek in 2023, with their founding years all in one graphic. It's a solid snapshot of over 175 years of business history, covering everyone from Heineken and Nokia to Tesla,…
DeepSeek v4 Flash is genuinely impressive. It's a 284B parameter model that outdoes GPT 5.6 Luna in real-world use, which feels unfair at that size. Sure, it's benchmaxxed like everything else, but the…
Do you realize what Moonshot just released? Kimi K3. 2.8 trillion parameters. The largest open model ever shipped. DeepSeek V4-Pro sits at 1.6T. Everything else this year was under 1T. 1 million token context, native,…
Kimi K3 2.8T parameters
Anthropic's CEO argues that open-source AI is a red herring, stating he doesn't care if a model is open but rather if it's better at the things that matter. He explains that in AI, you see the weights, not the source…
Anthropic CEO on open-source AI
This is the moment Chinese AI outperformed American AI. One of the biggest publicly traded crypto companies has abandoned OpenAI and Anthropic. Coinbase moved to open-weight Chinese models from Zhipu and DeepSeek,…
Coinbase switches to Chinese AI models
Open source has overtaken American labs in market share. On OpenRouter, the share of tokens used by US models collapsed from 73% to 33% in just one year. Meanwhile, Chinese open-weight models like DeepSeek have surged.…
GLM 5.2 has surged on the Artificial Analysis Coding Index, jumping from 43.4 to 50.7 , a gain of seven points in a single release. According to bridgemindai, these results align with what they're observing on…
Claude's Fable 5 built a horror game using the same prompt we gave Fusion. Don't be fooled by OpenRouter's misleading marketing. Fusion's performance doesn't come close to Fable 5. Combining Opus 4.8, GPT 5.5, and…
Claude Fable 5 horror game
