Explore all posts related to LLMs
Shanghai AI Lab's InternS2-Preview-397B is a 397B open-weight MoE model that tops MMLU-Pro at 89.75 — beating GPT-5.5 — and leads open-source general reasoning, with specialized training across 20+ scientific domains.
Elon Musk has confirmed that xAI's next model — a 2-trillion-parameter system — already outperforms Grok 4.5 on every internal benchmark, with initial training expected to finish within the week.
Kimi K3's 2.8 trillion parameters get all the attention — but the capabilities most users haven't discovered yet are what make Moonshot AI's new flagship genuinely worth trying.
A developer used a single prompt to get Kimi K3 to generate a fully functional Window Browser OS — and the demo is going viral for what it implies about AI's one-shot generative ceiling.
TokenRouter is running GLM-5.2 on a free shared tier through July 25 — and its full-context paid tier at 60% off, stackable to 72% off list with the Top Up bonus. Here's exactly what each tier gets you and what the fine print says.
Max Weinbach used Kimi K3's agent swarm to build a fully interactive macOS 27 simulator — with Liquid Glass effects, playable Chess, and floating widgets — in a matter of hours. Here's what it reveals about the model's frontend capabilities and what it means for developers.
