📊 Full opportunity report: The Weight Of Evidence: What Thinking Machines’ Inkling Tells Us on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has publicly released the full weights of its new multimodal AI model, Inkling, under an open license. This move emphasizes transparency and ownership, but raises questions about restrictions and data sources.
Thinking Machines has released the full weights of its newest AI model, Inkling, under the Apache 2.0 license, making it freely downloadable and modifiable. This is notable because the company chose open distribution over a closed API, emphasizing transparency and ownership for users. The release was made available on Hugging Face, along with support for multiple frameworks, marking a significant shift in how large AI models are shared and used in the industry.
Inkling is a 975-billion-parameter multimodal transformer, supporting text, images, and audio inputs, with a 1-million-token context window. It was trained on 45 trillion tokens across various modalities, with a unique encoder-free design for multimodal input processing. The model’s weights are openly available under the Apache 2.0 license on Hugging Face, allowing users to download, modify, and deploy independently.
In addition to the flagship, a smaller version, Inkling-Small (276B parameters), was also previewed, showing competitive performance on several benchmarks. The training process involved hybrid optimizers and over 30 million reinforcement learning rollouts, with some training data generated by open-weight models like Kimi K2.5. The company explicitly states that while the weights are open, the training data and pipeline are not publicly disclosed.
Critically, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP), which may impose restrictions on surveillance, deception, and automated decision-making, potentially conflicting with the open-source license. The company has not publicly verified the AUP’s full text, but its existence highlights ongoing debates about model transparency versus usage 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 Release and Usage Policies
The release of Inkling’s full weights under an open license signifies a shift toward greater transparency and ownership in large AI models, enabling organizations to fine-tune, inspect, and deploy independently. However, the potential overlay of usage restrictions through a separate policy raises questions about the true openness of the model. This development could influence industry standards around open-source AI, especially regarding licensing, data privacy, and ethical use.
For developers and organizations, the ability to own and modify the model directly offers increased control, especially relevant after recent incidents of model shutdowns by authorities. Conversely, the existence of usage policies may limit certain applications, particularly in sensitive domains like surveillance or automated decision-making, where restrictions could impact deployment strategies.
LOCAL LLM DEPLOYMENT: Training, Fine-Tuning, & Offline Inference: The Complete Developer’s Guide to Building, Training, and Running Private Open-Source AI Offline (with full source code)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Trends Toward Transparency and Ownership
The release of Inkling follows a broader industry trend toward open-sourcing large language models, with companies increasingly opting to share weights openly to foster innovation and transparency. Historically, many models have been distributed with closed licenses or API-only access, limiting control and customization. Recent examples, including Meta’s Llama and other open-weight models, reflect a shift driven by community demand and regulatory considerations.
Thinking Machines, founded by a former OpenAI CTO and staffed with experts involved in ChatGPT’s development, has taken a notably transparent approach by releasing weights immediately and openly. This contrasts with some industry players that prefer API-based access to retain control, highlighting a divergence in strategies around model distribution and governance.
The controversy over licensing and usage restrictions underscores ongoing debates about the balance between openness, safety, and commercial interests in AI development.
“We believe in giving users the freedom to own and modify their models, but we also maintain responsible use policies to ensure ethical deployment.”
— Thinking Machines spokesperson
Unresolved Questions About Model Restrictions and Data Sources
It remains unclear how the separate Model Acceptable Use Policy (AUP) will be enforced and how it might limit certain uses of Inkling, especially in sensitive domains. The full text of the policy has not been publicly verified. Additionally, the specifics of the training data, including proprietary or sensitive sources, are not disclosed, raising questions about data transparency and potential biases.
Further, it is not yet confirmed whether the open weights include all training artifacts or if some components are restricted, which could influence how users interpret and deploy the model.
Next Steps for Model Adoption and Policy Clarification
Expect further disclosures from Thinking Machines regarding the full text of the AUP and detailed documentation on data sources. Industry observers will likely conduct independent benchmarks and safety assessments to verify claims about Inkling’s performance and safety features.
Additionally, organizations interested in deploying Inkling will evaluate the licensing terms and restrictions before integrating it into their systems. The broader AI community will watch how the model’s open release influences industry standards around transparency, ownership, and responsible use.
Key Questions
What makes Inkling different from other large language models?
Inkling is a multimodal transformer with 975 billion parameters, supporting text, images, and audio inputs, and is openly available under the Apache 2.0 license—offering full ownership and customization options.
Does open release mean the model can be used freely in all applications?
While the weights are openly available, reports suggest there may be separate usage restrictions through a policy. Users should verify the full terms before deploying in sensitive or regulated contexts.
Why is the licensing and policy distinction important?
The Apache 2.0 license permits modification and commercial use, but if a separate policy imposes restrictions, it could limit certain applications or require compliance checks, impacting how freely the model can be used.
What are the risks of using an open-weight model with restricted policies?
Potential risks include unintended restrictions on deployment, legal uncertainties, or ethical concerns if the policy limits certain types of use, especially in sensitive areas like surveillance or automation affecting human rights.
What will happen next in the development of Inkling?
Further transparency from Thinking Machines regarding the full policy and training data, along with independent benchmarking, will shape how the industry adopts and regulates open models like Inkling.
Source: ThorstenMeyerAI.com
