📊 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 has published the full weights of its Inkling model under Apache 2.0 on Hugging Face, making it openly available. This move signals a shift toward transparency, but the company maintains a separate use policy that could limit how the model is used, raising questions about true openness.
Thinking Machines has publicly released the full weights of its new Inkling model on Hugging Face under the Apache 2.0 license, marking a notable shift in AI model transparency. Unlike typical launches that restrict access or offer only API endpoints, this release provides the complete model weights from day one, allowing for independent inspection, modification, and deployment. This development comes amid ongoing debates about open-source AI and the implications of owning versus renting models, making it a significant event for the AI community and potential users.
The Inkling model is a 975 billion-parameter mixture-of-experts transformer supporting a 1-million-token context window. It was trained on 45 trillion tokens across text, images, audio, and video, and features a multimodal, encoder-free design capable of processing inputs from multiple modalities—text, images, and audio—jointly. The weights are now available openly on Hugging Face, with full licensing under Apache 2.0, which permits download, modification, and commercial use.
However, the announcement also revealed that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP), which restricts certain applications such as surveillance, deception, and fully automated decision-making affecting individuals’ rights. This layered policy contrasts with the permissiveness of the Apache 2.0 license and raises questions about the true openness of the model. The company also disclosed that the training data and pipeline are not publicly available, adhering to industry norms but limiting full transparency.
Furthermore, the model’s performance has been benchmarked externally, with strengths in speech and adversarial safety but middling results in some natural language understanding tasks. The full weights are now accessible, but the licensing and use restrictions remain key points of debate among observers and potential users.
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 and Use Policies
This release marks a shift toward greater model transparency in the AI industry, enabling organizations to own and control their models without dependency on API providers. It also challenges the traditional notion of open source by combining open weights with a restrictive use policy, highlighting ongoing tensions between openness and control. For industries like public safety, geospatial analysis, and critical infrastructure, these developments could influence how models are adopted and regulated.
However, the layered restrictions could limit the model’s practical deployment and raise concerns about enforceability. The move also signals a possible shift in industry norms, where openness is paired with specific use restrictions rather than unconditional freedom, affecting future AI licensing and community standards.
RayNeo Air 4 Pro AR Glasses & JoyDock Bunlde
RayNeo AR big-screen gaming: Connect your JoyDock to Switch and dive into immersive gaming. Enjoy a theater-sized 201-inch…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Model Transparency and Industry Norms
Historically, most large language models have been released with API access only, or with weights under restrictive licenses that prevent modification or commercial use. Open-source efforts like GPT-Neo and LLaMA have aimed for transparency but often limited their licensing or access. Recently, some companies have begun releasing models with open weights, but often with layered restrictions or separate policies governing use.
Thinking Machines, founded by former OpenAI CTO, has built a reputation for transparency and rigorous benchmarking. Its decision to release Inkling’s full weights openly under Apache 2.0, while maintaining a separate use policy, represents a nuanced approach that emphasizes ownership and control. This move follows a broader industry conversation about balancing openness with safety and ethical considerations, especially as models grow larger and more capable.
“Our goal is to enable owners to deploy and adapt Inkling freely while ensuring responsible use through our policies.”
— Thinking Machines spokesperson
ESP32 Basic Starter Ai Chatbot Kit Development Board USB-C Dual Core Microcontroller Support AP/STA/AP+STA Compatible with Arduino IDE IoT for Beginners Engineers
【High Performance】The ESP32 module is equipped with a dual-core CPU and features a Type-C USB interface, with 44…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Licensing and Enforcement
It remains unclear how strictly the Model Acceptable Use Policy will be enforced, and whether it will limit the practical use of the open weights. The scope of restrictions, their legal enforceability, and how they will impact commercial or research applications are still under discussion. Additionally, the extent of transparency regarding training data and pipeline remains undisclosed, leaving gaps in full understanding of the model’s development and potential biases.
Building A large language model with Ai: A Practical Guide to Structuring LLM Systems from Scratch Using Reverse-Engineering Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Policy Clarification
Organizations and developers will likely begin testing Inkling’s weights in various applications, scrutinizing the licensing and use policies. Further transparency about the training data and pipeline may follow, along with potential clarifications from Thinking Machines regarding enforcement of restrictions. Industry observers will monitor how this layered approach influences future model releases and licensing norms.
AI model licensing compliance tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What does the open release of Inkling weights mean for AI transparency?
It allows organizations to inspect, modify, and deploy the model independently, promoting greater transparency and control compared to API-only models.
How does the Model Acceptable Use Policy affect the openness of Inkling?
While the weights are open under Apache 2.0, the separate use policy imposes restrictions that could limit certain applications, complicating the narrative of full open-source access.
Will the training data or pipeline be released?
No, the training data and pipeline are not publicly disclosed, which limits full transparency into the model’s development process.
What are the potential risks of layered licensing restrictions?
They could lead to inconsistent enforcement, limit practical deployment, and create legal uncertainties for users relying on the open weights.
What is the significance of this release for the AI industry?
It signals a possible shift toward more open models with layered restrictions, influencing licensing practices and community standards in AI development.
Source: ThorstenMeyerAI.com
