📊 Full opportunity report: The Subtle Significance Of Thinking Machines’ Inkling In AI Development on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released its Inkling model, a 975-billion-parameter open-weight transformer, openly available on Hugging Face under Apache 2.0. The release highlights transparency and raises questions about licensing and use restrictions. The development marks a significant step in open AI model deployment and ownership.
Thinking Machines Lab has announced the release of its first foundation model, Inkling, a 975-billion-parameter multimodal transformer, available openly on Hugging Face under Apache 2.0. This marks a shift in AI model deployment, emphasizing transparency and ownership, and directly addresses ongoing industry debates about licensing and control.
The Inkling model is a mixture-of-experts transformer supporting a one-million-token context window, pretrained on 45 trillion tokens across text, images, audio, and video. It was trained using a hybrid optimizer on NVIDIA systems, with over 30 million reinforcement learning rollouts, and was developed with an unusually candid approach to its training process and specifications.
Importantly, the weights are released under Apache 2.0 license, allowing download, modification, and deployment on private infrastructure. However, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP) that restricts surveillance, deception, and automated decision-making affecting individuals’ rights, raising questions about the true openness of the release.
While the model performs strongly on benchmarks like AIME 2026 and VoiceBench, it shows middling results on others such as Humanity’s Last Exam and Terminal-Bench, reflecting a balanced but not top-tier performance across all tasks. The release also included a smaller variant, Inkling-Small, which matches or surpasses larger models on several benchmarks thanks to an improved training recipe.
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-Weight Model Release
The release of Inkling under an open license represents a significant milestone in AI development, emphasizing transparency, ownership, and control over powerful models. This contrasts with typical commercial API-based models, which restrict access and modification. The move could influence industry standards on open-source AI, encouraging more companies to release models with fewer restrictions, thereby fostering innovation and competition.
However, the potential layered restrictions via the AUP introduce questions about the true openness and whether the model can be freely used in sensitive domains such as surveillance or public safety. The debate highlights ongoing tensions between openness, safety, and commercial interests in AI development.
For developers and organizations, owning a model like Inkling means greater flexibility and control but also greater responsibility for managing ethical and legal considerations, especially if restrictions are layered on top of open licenses.
Accelerate Everything with Tensor Cores: A Developer’s Guide to High-Performance AI, Efficient Training, and Scalable Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Trends Toward Open AI Models
Over the past year, a growing number of AI labs have begun releasing large models with open weights, aiming to foster transparency and democratize AI access. Notable examples include Meta’s Llama 2 and Stanford’s Alpaca, which have set precedents for open licensing. Thinking Machines’ approach — releasing a large, multimodal model with full weights and openly sharing training details — aligns with this trend but also introduces new questions about licensing and restrictions.
Historically, many companies have preferred API-based models to retain control and monetize their offerings, but the open model movement challenges this paradigm, emphasizing ownership and customization. The debate continues over how to balance openness with safety and ethical considerations, especially with models capable of complex reasoning and multimodal processing.
The release of Inkling adds to this evolving landscape, highlighting both the opportunities and challenges of open AI development in a rapidly shifting industry environment.
“We believe in empowering users with full ownership of their models while maintaining responsible use policies.”
— Thinking Machines spokesperson
multimodal AI development tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Inkling’s Open License and Use Restrictions
It remains unclear how the Model Acceptable Use Policy (AUP) will be enforced and whether it will effectively limit certain applications despite the open weights license. The exact scope, enforceability, and legal weight of this policy are still under question.
Additionally, the long-term impact of this release on industry standards and whether other labs will follow suit with similarly open models remains to be seen. The full testing and independent benchmarking of Inkling are ongoing, and its real-world performance and safety profile are yet to be fully assessed.
open-source AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Industry Adoption and Model Evaluation
Expect further independent testing and benchmarking of Inkling’s performance and safety. The community will scrutinize the AUP and licensing restrictions to determine their practical enforceability and impact.
Other AI labs may follow with their own open releases, potentially shifting the industry toward more transparent and ownership-oriented models. Meanwhile, organizations interested in deploying Inkling will evaluate its capabilities, restrictions, and compliance with their ethical standards.
Regulators and policymakers may also monitor this development as it influences debates on AI safety, ownership, and regulation in the era of large-scale open models.
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
Versatile Licensing Options: PCmover Business offers flexible licensing with Standard License tiers for 1, 5, 10, or 25…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What makes Inkling different from other large language models?
Inkling is a 975-billion-parameter multimodal transformer released openly under Apache 2.0, supporting text, images, and audio inputs, with full weights available for download and modification.
Does the open license mean the model can be used freely in all applications?
While the weights are openly available, reports suggest there may be additional restrictions via a separate Acceptable Use Policy, which could limit certain applications, especially those involving surveillance or automated decision-making.
Why is the layered restriction on the model significant?
It raises questions about the true openness of the release, as the Apache 2.0 license alone does not impose such restrictions, and enforcement of the AUP could limit how the model is used in practice.
What are the potential impacts of this release on the AI industry?
The release signals a shift toward more open and ownership-focused models, potentially encouraging more labs to share powerful models openly, but also prompting debates about safety, licensing, and ethical use.
Source: ThorstenMeyerAI.com
