July 15, 2026
Open weights, open season
Inkling: Our Open-Weights Model
AI fans are cheering, roasting, and asking if this is America’s answer to China
TLDR: Inkling released an open AI model people can freely customize, a notable move in a market where many systems stay locked down. Fans are calling it a much-needed Western challenger, while skeptics are already asking why they should care if rivals still seem stronger.
Inkling just did the thing skeptics kept joking it never would: actually release a model. The company says its new AI can handle words, pictures, and audio, remembers a huge amount of context, and is being handed out with the full weights so developers can tweak it themselves. In plain English: this is a big “here, go build with it” moment, not just another glossy promise. And yes, the community noticed. One commenter basically summed up the shock with, “I never thought I’d see the day they released a model, rather than a blog post” — which is both a compliment and a roast.
The comment section quickly turned into a mini geopolitical cage match. One of the loudest reactions was that the US badly needs an open AI model that can compete with popular Chinese ones, with some users openly saying they’ve been rooting for Chinese models because there hasn’t been a strong Western alternative. That made Inkling feel bigger than just a product launch: to some fans, it’s a potential comeback story.
But not everyone is swooning. Critics immediately dragged out the benchmark receipts, asking why anyone should bother if it’s bigger than rivals like GLM yet still not clearly better. Others landed somewhere in the middle: great at following directions, maybe less exciting at coding, but still valuable because variety matters. The funniest jab? A user clowning on a demo that looked like a humble Chrome window on localhost — a wonderfully unglamorous image for a company trying to launch its AI era with flair.
Key Points
- •The article announces Inkling, an open-weights multimodal AI model released with full weights available for customization.
- •Inkling is described as a mixture-of-experts transformer with 975B total parameters, 41B active parameters, a 1M-token context window, and pretraining on 45 trillion tokens across text, images, audio, and video.
- •Thinking Machines says Inkling is designed for broad capabilities across agentic tasks, reasoning, coding, instruction-following, factuality, vision, and audio rather than state-of-the-art performance in a single domain.
- •Inkling is available for fine-tuning on Tinker, and the company introduced an Inkling Playground in the Tinker console for developer interaction.
- •The article highlights demos and evaluations including one-shot web app creation, embedded browser use, and a 40-iteration snake game refinement process reviewed by GPT Codex.