July 15, 2026
Big AI, bigger comment drama
Murati's Thinking Machines Releases Open-Weights 975B Parameter LLM
A giant new AI dropped, and the crowd instantly argued over the numbers and whether this was old news
TLDR: Thinking Machines released a massive new open AI model that can handle words, images, and sound, and people can fine-tune it themselves. But the comments quickly turned into a squabble over whether the giant size claim really counts — and whether the post was already old news.
Mira Murati’s Thinking Machines just unveiled a huge open AI model called Inkling, promising a system that can work with text, pictures, and audio and can be downloaded and customized by others. On paper, that’s a big deal: a rare mega-scale release that invites outsiders to tinker with it instead of keeping everything locked behind a corporate curtain. But if you were expecting the comment section to burst into applause, think again — the real energy was pure internet nitpicking.
The strongest reaction was a classic numbers fight. One commenter immediately zoomed in on the model card and dropped the eyebrow-raising stat: “975B parameter 41B active”. In plain English: the model is being marketed as enormous, but only part of it is actually working at one time — and that detail became the thread’s instant obsession. Was this impressive efficiency, or flashy label math? The comments didn’t really settle it; they just gave off that familiar “okay, but what’s the real size?” energy.
Then came the second wave of drama: the dupe police. Another user declared the post already discussed elsewhere, turning the launch into a miniature replay of one of the internet’s oldest traditions — not celebration, but arguing over whether everyone is late to the party. One comment was even flagged, which only adds to the delicious sense that this release landed with less of a standing ovation and more of a chaotic, half-skeptical group chat vibe.
Key Points
- •Thinking Machines released an open-weights large language model.
- •The model is described as efficient.
- •The model supports text, image, and audio modalities.
- •The model is available for fine-tuning.
- •The article focuses on the model's accessibility and multimodal capabilities.