📊 Full opportunity report: How To Interpret Thinking Machines’ Inkling For AI Trends on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines released its first foundation model, Inkling, with open weights on Hugging Face under Apache 2.0. The model is not the strongest but emphasizes transparency and ownership, signaling shifts in AI development.
Thinking Machines has released the full weights of its first foundation model, Inkling, on Hugging Face under an Apache 2.0 license. This move emphasizes transparency and ownership, contrasting with typical industry practices of closed models, and signals a shift in how AI models are distributed and used.
The Inkling model is a 975-billion-parameter Mixture-of-Experts transformer supporting multimodal inputs—text, images, and audio—trained on 45 trillion tokens. It was released with open weights, allowing users to download, modify, and deploy independently, which is a departure from most proprietary models. The release includes a smaller version, Inkling-Small, which performs competitively on several benchmarks, indicating promising capabilities.Despite the open release, there are important caveats. The weights are under Apache 2.0, but the training data and pipeline are not publicly available. Additionally, reports suggest Thinking Machines maintains a separate Model Acceptable Use Policy that restricts surveillance, deception, and automated decision-making, raising questions about the true openness of the model. The company’s approach highlights a focus on transparency and user ownership, but the presence of restrictions layered on top of the open license complicates the narrative. Industry observers note that such layered policies could influence how organizations adopt and trust the model, especially in sensitive domains like public safety or geospatial analysis.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 for AI Ownership and Open-Source Models
This release marks a notable shift in AI development, emphasizing model ownership and transparency over proprietary control. By providing open weights under Apache 2.0, Thinking Machines enables organizations to fine-tune, inspect, and deploy the model independently, potentially reducing reliance on cloud providers and closed ecosystems.
However, the layered use restrictions introduce complexity, as they may limit how the model can be applied in certain contexts. This approach could influence industry standards, encouraging more companies to balance openness with responsible use policies, especially amid increasing regulatory scrutiny and societal concerns about AI misuse.
The Practical Guide to Large Language Models: Hands-On AI Applications with Hugging Face Transformers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Open Models and Industry Shifts
Over the past year, there has been a growing debate around the openness of foundational AI models. Major players like OpenAI and Meta have typically released models with closed weights or limited access, citing safety and commercial concerns. In contrast, some organizations, including Stability AI and now Thinking Machines, are experimenting with open-release strategies.
Thinking Machines, founded 17 months ago by former OpenAI CTO, has built models like Inkling with a focus on transparency and user control. The recent release aligns with broader industry discussions about democratizing AI access, but it also raises questions about responsible use and potential misuse.
“We believe in empowering users with ownership and transparency, but we also prioritize responsible AI deployment through our Acceptable Use Policy.”
— Thinking Machines spokesperson
Unclear Aspects of Inkling’s Open-Source Nature
It remains unclear how enforceable the Model Acceptable Use Policy is in practice, and whether it effectively limits the model’s application despite the open weights. The full training data, pipeline, and detailed safety measures have not been disclosed, raising questions about the true level of transparency and openness.
Additionally, the impact of layered restrictions on commercial use and third-party modifications is still developing, and industry observers await independent assessments and real-world deployments to better understand these dynamics.
Next Steps in Adoption and Industry Impact
Organizations and developers will likely begin experimenting with Inkling’s open weights, testing its capabilities and limitations across various applications. Monitoring how the layered use restrictions are enforced and how the community responds will be critical.
Further releases, transparency disclosures, and independent evaluations are expected to clarify the model’s openness and safety profile. Industry-wide, this approach could influence future model releases, balancing openness with responsible use policies.
Key Questions
What makes Inkling different from other foundation models?
Inkling is released with open weights under Apache 2.0, allowing independent use, modification, and deployment. It also supports multimodal inputs and has a large parameter count, but layers restrictions via a separate Acceptable Use Policy.
Can organizations freely modify and commercialize Inkling?
Under the Apache 2.0 license, organizations can modify and commercialize the model. However, reports suggest a separate use policy may impose restrictions, so clarity on this point is essential before deployment.
Why is the layered use policy significant?
The policy could limit how the model is used in sensitive applications, creating a tension between open licensing and responsible deployment, especially in areas like surveillance or automated decision-making.
Will the full training data and pipeline be released?
No, only the model weights are publicly available. The training data, pipeline, and safety measures remain undisclosed, which limits full transparency.
What does this release mean for the AI industry?
It signals a potential shift toward more open models that still include restrictions, influencing how companies balance transparency, ownership, and safety in AI development.
Source: ThorstenMeyerAI.com
