📊 Full opportunity report: Deciphering AI’s Future Through Thinking Machines’ Early Clues on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large open-weight AI model, openly sharing its weights and specifications. This move signals a shift towards transparency and ownership in AI development, but questions remain about licensing and data use.
Thinking Machines, a 17-month-old AI lab founded by former OpenAI CTO, has publicly released the full weights of its new foundation model, Inkling. This is notable because most large models are either closed or released with restrictions, but Inkling’s weights are available under the Apache 2.0 license, allowing broad use and modification. This development provides a rare glimpse into the inner workings of a major AI model and signals a potential shift in industry transparency and ownership practices.
The Inkling model is a Mixture-of-Experts transformer with 975 billion total parameters, supporting a 1-million-token context window. It was trained on 45 trillion tokens across text, images, audio, and video, with a native multimodal input design that processes text, images, and audio jointly, without relying on vision adapters. The model’s weights are now publicly available on Hugging Face under Apache 2.0, enabling anyone to download, modify, and deploy it independently.
In addition to the full model, Thinking Machines announced a smaller version, Inkling-Small, with 276 billion parameters and 12 billion active, which reportedly matches or surpasses its larger sibling on several benchmarks. The training process involved hybrid optimization and over 30 million reinforcement learning rollouts, with the model showing promising reasoning performance improvements. The company also disclosed that synthetic data from open models like Kimi K2.5 was used during training, highlighting openness in their approach.
However, the release is accompanied by a Model Acceptable Use Policy (AUP) that restricts surveillance, deception, and automated decision-making impacting individuals, raising questions about the true openness of the model despite its open weights. Critics note that the Apache 2.0 license does not impose such restrictions, and the layered policy could limit practical use or enforceability.
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’s full weights under an open license marks a significant shift towards transparency in AI development, allowing developers and organizations to own, fine-tune, and deploy the model independently. This could accelerate innovation, reduce reliance on proprietary APIs, and foster a more open AI ecosystem. However, the accompanying use restrictions and lack of disclosed training data complicate the narrative, raising questions about what “open” truly means in this context.
For industry watchers, this move signals a potential trend toward more openly accessible foundational models, challenging the dominance of closed, API-only offerings. For regulators and policymakers, it underscores the importance of clarifying licensing and usage policies to ensure responsible deployment and compliance with ethical standards.
Introduction To Open Source Ai Development With Ollama
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Trends in Open AI Models
Over the past year, the AI community has seen increasing calls for transparency and ownership, especially after incidents where large models were abruptly shut down or restricted by corporate or governmental actions. Historically, most large models have been released with limited access or closed licenses, making full transparency rare. The release of models like Meta’s Llama and now Thinking Machines’ Inkling under open licenses reflects a broader push for democratization of AI technology.
Previous efforts, such as OpenAI’s GPT-2 release and Meta’s Llama, set precedents for open access, but often with restrictions or delayed full releases. Inkling’s approach — releasing full weights immediately and openly — may influence industry standards, prompting other labs and companies to reconsider their release strategies amid increasing demand for transparency and control.
“Our goal is to empower the community with full access to our models, fostering innovation and responsible use.”
— Thinking Machines spokesperson
Advanced Language Tool Kit: Teaching the Structure of the English Language
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Questions About Licensing and Data Use Restrictions
While the weights are openly available, reports suggest that Thinking Machines maintains a separate Acceptable Use Policy that restricts surveillance, deception, and certain automated decisions. The scope and enforceability of this policy are not fully verified, raising questions about the true openness of the model and the potential for restrictions beyond the Apache 2.0 license. It remains unclear how these restrictions will impact practical deployment and whether they align with the open-source spirit.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Industry Adoption and Oversight
Expect further independent testing and benchmarking of Inkling’s smaller variant, Inkling-Small, once full weights are released. Industry observers will scrutinize the model’s compliance with the stated use restrictions and evaluate its performance across various domains. Regulatory bodies may also assess the implications of open-weight releases for responsible AI deployment. Additionally, other AI labs might follow suit, increasing the prevalence of openly accessible foundational models.
In parallel, ongoing discussions about licensing clarity, ethical use, and data transparency are likely to intensify, shaping the future landscape of AI ownership and control.
AI Engineering and Agentic AI: Designing Autonomous Language Model Systems with Memory, Tools, and Safe Deployment
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 AI models?
Inkling is notable for its full open-weight release under Apache 2.0, allowing unrestricted download, modification, and deployment, unlike most models which are either closed or have restricted access.
Does the open release mean the model is completely open-source?
No. The weights are openly available, but reports suggest there are additional restrictions via a separate Acceptable Use Policy, which could limit how the model is used in practice.
What are the potential risks of releasing such a large model openly?
Open access could lead to misuse, such as surveillance or automated deception, especially if restrictions are not enforceable or clearly communicated. It also raises concerns about data privacy and ethical deployment.
Will this influence other AI companies to release their models openly?
It could set a precedent, encouraging more labs to share models openly, but industry-wide adoption will depend on legal, ethical, and strategic considerations.
What should I watch for in upcoming developments?
Look for independent benchmarks of Inkling-Small, clarity on the enforceability of usage restrictions, and industry responses to this new open model approach.
Source: ThorstenMeyerAI.com
