Thinking Machines Lab, led by ex-OpenAI CTO Mira Murati, unveils Inkling, a 975B parameter model. It's a contender in the U.S., yet trails China's best.
Thinking Machines Lab is back with a bang. Former OpenAI CTO Mira Murati just introduced Inkling, a jaw-dropping 975 billion parameter model. This isn't just a tale of numbers, though. Inkling is shaking up the U.S. open-weights scene, leading the Artificial Analysis Intelligence Index stateside. But here's the rub: despite its impressive stats, it lags behind top models from China in some tasks.
Price and Positioning #
At $1.87 per million input tokens, Inkling isn't the cheapest date in town. Still, it's not trying to be the most powerful model either. Thinking Machines is pitching it as an ideal starting point for fine-tuning. That means they're banking on users who want to take a solid base and make it their own. Open weights don't wait for permission, right?
Why This Matters #
So why should you care? Well, with open weights and a massive parameter count, Inkling is primed for customization. The real action happens when you get to tweak and tune it to your needs. If you're not running it locally yet, you're late. This is a model that demands hands-on exploration.
The Bigger Picture #
Let's step back for a second. While Inkling is a headline-grabber in the U.S., it's hard to ignore the fact that Chinese models still outperform it on select metrics. This raises a question: in the race for AI supremacy, can the U.S. keep pace with China's relentless advancement? The speed difference isn't theoretical. You feel it. But with models like Inkling, the gap might just start to close.
For now, Inkling is a solid step forward for U.S. open-weights. It's a reminder that innovation in AI isn't slowing down. Another week, another open model doing what the big labs promised. Get AI news in your inbox
Daily digest of what matters in AI.
Key Terms Explained #
Fine-Tuning The process of taking a pre-trained model and continuing to train it on a smaller, specific dataset to adapt it for a particular task or domain.
OpenAI The AI company behind ChatGPT, GPT-4, DALL-E, and Whisper.
Parameter A value the model learns during training — specifically, the weights and biases in neural network layers.