📊 Full opportunity report: The Role Of Weights In AI: Insights From Thinking Machines’ First Clues on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has made the weights of its Inkling model publicly available under Apache 2.0. This move is significant for transparency and control, but raises questions about licensing and restrictions.
Thinking Machines has officially released the open weights of its latest foundation model, Inkling, on Hugging Face under the Apache 2.0 license. This marks a notable shift in AI model distribution, emphasizing openness and control for users, and directly addressing ongoing debates about model ownership and licensing in the industry.
Inkling is a multimodal, 975-billion-parameter transformer trained on 45 trillion tokens of text, images, audio, and video. It supports a 1-million-token context window and was trained with a hybrid optimizer on NVIDIA systems. The full weights are now publicly available, with day-zero support in multiple open-source frameworks, including transformers and llama.cpp.
The release is significant because it is the first time in recent memory that a major AI lab has openly published the complete model weights alongside candid training details. Unlike many industry releases, which often restrict access or keep models closed, Inkling’s weights are licensed under Apache 2.0, allowing download, modification, and commercial use, with some caveats.
However, there are important limitations: the training data and pipeline are not published, and reports suggest that Thinking Machines enforces a separate Model Acceptable Use Policy (AUP) that restricts surveillance, deception, and automated decision-making affecting rights. This layered policy raises questions about the true openness of the model and its intended use cases.
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 Licensing Restrictions
This move signals a shift toward greater transparency and user control in AI model distribution, potentially setting a new industry standard. The open release allows organizations to fine-tune, inspect, and deploy Inkling independently, reducing reliance on proprietary APIs and closed models.
At the same time, the layered restrictions via the AUP complicate the narrative of open-source AI, raising questions about enforceability and scope. For sectors like public safety or geospatial analysis, these restrictions could influence whether the model is usable or legally deployable.
Overall, this development underscores ongoing tensions between openness, control, and responsible use in AI, with industry-wide implications for how models are shared and governed.
Samsung 50-Inch Class Crystal UHD U8000H Series Samsung Vision AI Smart TV (2026 Model, 50U8000H) Crystal Processor 4K, Endless Free Content, Motion Xcelerator, Color Booster, Alexa Built-in
- Crystal Processor 4K: Enhances colors and sharpens details
- Endless Free Content: Access 2,700+ free streaming options
- Samsung TV Plus: 750+ subscription-free channels included
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Model Releases and Industry Norms
In recent years, most AI labs have favored controlled access to their models, often releasing only APIs or limited weights, citing safety and misuse concerns. Open-source efforts like Meta’s Llama and EleutherAI’s models have pushed for more transparency, but full open weights remain rare among the largest models.
Thinking Machines, founded by former OpenAI CTO and staffed with talent from ChatGPT’s development, has adopted a different approach. Its previous projects emphasized transparency, but the full release of Inkling’s weights under an open license marks a notable departure from the norm, aligning with a broader push for open AI ecosystems.
Prior to this, industry leaders have been cautious about releasing models openly, citing risks of misuse and safety. This release, therefore, stands out as a deliberate statement about openness, balanced with layered restrictions through the AUP.
“Our goal is to enable responsible use while providing full access to our model’s weights for research and development.”
— Thinking Machines spokesperson
AI Robotic Arm Kit Hiwonder SO-ARM101 Embodied Imitation Learning Open Source 6-Axis Robot Arm 12 High-Torque Bus Servo Motors AI Vision Recognition (Standard Kit, Included 3D Printed Part, Assembled)
- Imitation Learning Platform: Supports end-to-end AI imitation learning
- Dual-Camera System: Includes gripper and external cameras for vision
- High-Torque Bus Servos: 12 magnetic feedback servo motors for smooth motion
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Model Use and Restrictions
It remains unclear how enforceable the Model Acceptable Use Policy (AUP) will be and whether it will significantly restrict commercial or research applications. The scope and legal weight of the layered restrictions are still being evaluated by industry observers.
Additionally, the impact of not releasing the training data or pipeline on transparency and reproducibility is uncertain. Critics may argue that true openness requires full disclosure of training sources and methods.
Finally, the long-term implications for industry standards and whether other labs will follow suit remain to be seen.
Hands-On Generative AI with Transformers and Diffusion Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Industry Adoption and Model Testing
Organizations and researchers will likely begin testing Inkling’s open weights, assessing performance across domains and safety benchmarks. Independent verification of claims and the enforceability of restrictions will be critical in the coming months.
Further updates are expected from Thinking Machines regarding the release of the full training data, pipeline details, and clarification of the AUP’s scope. Industry watchers will also monitor whether other labs adopt similar open release strategies or maintain proprietary controls.
Regulatory and legal discussions about layered licensing and open-source standards are also anticipated to shape future practices.
Laplink PCmover – Easy Migration of your Applications, Files and Settings from an Old PC to a New PC – Data Transfer Software with Optional Super Speed USB 3.0 Cable – Business Standard, 10 Licenses
- Flexible Licensing Tiers: 1, 5, 10, or 25 transfers
- Unlimited Transfer Attempts: One license per source-destination pair
- Suitable for Multiple PCs: Supports individual or multi-PC migrations
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is unique about Thinking Machines’ Inkling model release?
It is the first major AI model to release full open weights under Apache 2.0 license while explicitly stating it is not the most powerful model available and imposing additional restrictions via a separate AUP.
Does open weights mean the model is fully open source?
No. The weights are under Apache 2.0, but the training data and pipeline are not published. Also, the layered AUP introduces restrictions that complicate the open-source classification.
What are the main limitations of Inkling’s open release?
The training data and pipeline are not shared, and the Model Acceptable Use Policy restricts certain applications like surveillance and automated decision-making, which may limit how freely the model can be used.
How might this release influence the AI industry?
It could set a precedent for more transparent model sharing with layered restrictions, encouraging a balance between openness and responsible use, but also raising questions about enforceability and true openness.
What should users consider before deploying Inkling?
Users should review the licensing terms and the Model Acceptable Use Policy carefully, especially if their application involves sensitive areas like surveillance or automated decision-making.
Source: ThorstenMeyerAI.com
