📊 Full opportunity report: The Weights Came First: What Thinking Machines’ Inkling Actually Signals on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines publicly released the weights of its new multimodal model Inkling before the full model launch, emphasizing transparency and ownership. This marks a shift in AI deployment strategies, highlighting open access and honesty about model performance. Key details about licensing and use restrictions remain under scrutiny.
Thinking Machines has released the open weights of its latest multimodal foundation model, Inkling, on Hugging Face before unveiling the full model or API. This unconventional move signals a shift in AI deployment, emphasizing transparency and user ownership over proprietary control, which could influence industry standards.
The Inkling model is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active parameters, supporting a one-million-token context window. It was trained on 45 trillion tokens, including text, images, audio, and video, with a native multimodal, encoder-free design. The open weights are available under Apache 2.0 license, allowing download, modification, and commercial use, marking a departure from typical closed or API-only releases.
In addition to the full model, a smaller variant, Inkling-Small, with 276 billion total parameters, has been previewed. Its performance on various benchmarks suggests it can match or surpass larger competitors. The training process involved hybrid optimization and over 30 million reinforcement learning rollouts, with some data generated by open-weight models like Kimi K2.5, a Chinese model. The release’s transparency contrasts with industry norms, which often keep training data and pipelines proprietary.
However, reports indicate that Thinking Machines may impose a separate Model Acceptable Use Policy (AUP), restricting surveillance, deception, and automated decision-making affecting rights. This layered policy raises questions about the true openness of the release, as Apache 2.0 license alone does not impose such restrictions.
The weights came first: what Inkling actually signals
Mira Murati’s lab shipped its first foundation model — and the model isn’t the story. The order of operations is: full weights, Apache 2.0, day one, before any closed API. Plus a rare concession — the lab says it’s not the strongest model available, open or closed.
- AIME 2026 97.1%
- GPQA Diamond 87.2%
- MCP Atlas (Nemotron 44.7%) 74.1%
- VoiceBench · open-weight audio frontier 91.4%
- FORTRESS adversarial · best open 78.0%
- ForecastBench · calibration 61.1
- HLE text-only (GLM-5.2 40.1%) 29.7%
- SWE-bench Pro (GLM-5.2 62.1%) 54.3%
- Terminal-Bench 2.1 (GLM-5.2 82.7%) 63.8%
- SWE-bench Verified (Fable 5 95.0%) 77.6%
- Design Arena · 2nd open, behind GLM-5.2 ~10th
A 0.2 → 0.99 effort setting trades reasoning tokens against cost & latency, so you get a curve, not a point. On Terminal-Bench 2.1 it reportedly matches Nemotron 3 Ultra at ~⅓ the tokens. Peak score is a vanity metric when you serve millions of calls; the cost curve is what ships. (Bonus: its chain of thought compressed on its own during RL — nobody rewarded it; efficiency did.)
Pitched as the Western alternative to Chinese open weights (censorship-resistance training is the differentiator). But GLM-5.2 still wins on agentic/reasoning and Kimi K2.6 often on multimodal: best American open model, second in the open field. The irony — post-training was bootstrapped on synthetic data from Kimi K2.5.
BF16 needs ≥2 TB aggregate VRAM (8× B300 / 16× H200). NVFP4 still needs ≥600 GB. Not a workstation model — a 512 GB fleet falls just short. “Open” ≠ “runnable.” Mitigations: 1-bit GGUFs (~74% acc.), hosted eval routes, and Inkling-Small (12B active) — the release local-first builders actually want.
Open weights used to be a consolation prize. Inkling is a strategic open release — Apache 2.0, natively multimodal, honestly marketed, published complete on day one, optimized for deployment rather than headlines (the model isn’t the product; the fine-tuning platform is). It doesn’t need to win every benchmark for that to matter. The frontier is learning that owning the base beats renting the API — arriving now from the inside. For the sovereignty buyer: ① a real Western hedge against being switched off · ② verify the use policy before you build · ③ check the VRAM, then benchmark vs GLM-5.2 & Kimi K2.6 on your task.
Implications of Open Weights Before Full Model Launch
This approach signals a shift toward greater transparency and user ownership in AI development, potentially influencing industry standards. By releasing open weights first, Thinking Machines challenges the norm of API-only or closed models, emphasizing that owning and inspecting the model is valuable for developers and organizations. It also raises questions about how licensing and use restrictions are communicated and enforced, especially if layered policies exist.
For industries relying on AI transparency—such as public safety, geospatial analysis, and critical infrastructure—this move could set new expectations for openness and control. However, the potential restrictions via the AUP highlight ongoing tensions between openness and control, which users must scrutinize before adopting.
AI model weights download
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and the Significance of Open Weights
Historically, foundation models are released with APIs, closed weights, or limited access, primarily to control distribution and usage. Open-source releases under licenses like Apache 2.0 are rare, especially for models of Inkling’s scale. Recent industry shifts include some companies releasing weights openly but often accompanied by restrictive policies or limited data transparency. The decision by Thinking Machines to release weights first, openly, and honestly about the model’s performance and limitations, marks a notable departure from this trend.
Previous incidents, such as the worldwide shutdown of frontier models last month, have heightened awareness of the risks and responsibilities associated with model ownership. The move to release weights openly aligns with broader calls for transparency but also underscores the need for clear licensing and use policies, which remain under discussion.
“We believe in giving users full ownership and transparency of our models, starting with Inkling’s weights.”
— A spokesperson from Thinking Machines
HIWONDER Robot Car with ChatGPT Large AI Models, 3D Depth Camera Ackermann Chassis ROS2-HUMBLE Lidar SLAM Mapping Navigation Autonomous Driving, MentorPi A1 Standard Kit with Raspberry Pi5 8GB
For Raspberry Pi 5 & ROS2 Robot Car. MentorPi A1 smart AI robot car is powered by Raspberry…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Uncertainties Surrounding Licensing and Use Restrictions
It remains unclear whether Thinking Machines’ separate Model Acceptable Use Policy (AUP) imposes restrictions beyond the Apache 2.0 license. The details of this policy, including enforceability and scope, have not been publicly verified. This ambiguity raises questions about the true openness of the weights and what users can legally do with the model.
Fine-Tuning Large Language Models: From Custom Datasets to High-Performance AI Models Using Modern Toolchains
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps in Inkling’s Deployment and Policy Clarification
Expect further clarifications from Thinking Machines regarding the AUP and licensing details. The company may release detailed documentation or updates on permissible use, and independent testing will likely follow to verify performance claims. Additionally, other organizations may adopt or scrutinize this approach, influencing industry standards for model openness and control.
AI model licensing and use policies
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why did Thinking Machines release the weights before the full model?
To emphasize transparency and give users full ownership of the model, challenging the industry norm of API-only or closed releases.
Does open weights mean the model is fully open source?
No. The weights are licensed under Apache 2.0, but reports suggest there may be additional use restrictions via a separate policy, which complicates the notion of full openness.
What are the potential risks of releasing open weights first?
It could lead to misuse or unintended applications if restrictions are not clearly communicated or enforced, especially if layered policies limit certain uses despite open licensing.
How does this move impact the AI industry?
It could set a precedent for greater transparency and ownership, but also sparks debate over licensing, restrictions, and true openness in model deployment.
What should organizations do before adopting Inkling?
Carefully review the licensing terms and any associated use policies, and monitor for further clarifications from Thinking Machines.
Source: ThorstenMeyerAI.com
