📊 Full opportunity report: What We Can Expect Next In AI Based On Thinking Machines’ Inkling on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large open-weight foundation model with 975 billion parameters, openly available on Hugging Face under Apache 2.0. This move emphasizes transparency and ownership but raises questions about licensing and use policies.
Thinking Machines has released its first foundation model, Inkling, openly available on Hugging Face under the Apache 2.0 license. This marks a significant shift in AI model distribution, emphasizing transparency and ownership, and directly addresses ongoing debates about model access and control.
Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active, supporting a 1-million-token context window. It was pretrained on 45 trillion tokens across text, images, audio, and video, with a native multimodal input design that processes text, images, and audio jointly without additional vision adapters. The model’s weights are openly available on Hugging Face under Apache 2.0, enabling download, modification, and deployment by anyone, including commercial entities. The training involved hybrid optimization and over 30 million reinforcement learning rollouts, with some training data generated by open models like Kimi K2.5. Despite the openness of the weights, the company has reportedly maintained a separate acceptable use policy that restricts surveillance, deception, and automated decision-making affecting individuals, raising questions about the scope of openness and enforceability. Notably, the model outperforms many open-weight counterparts on several benchmarks, especially in speech and safety-related tasks, but scores mid-pack or behind on certain language understanding benchmarks.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 for AI Ownership
The release of Inkling as an openly available model under Apache 2.0 represents a shift toward greater transparency and control in AI development. It allows organizations to fine-tune, inspect, and deploy the model independently, reducing reliance on proprietary APIs. This move could accelerate innovation, foster competition, and influence industry standards for model licensing. However, the presence of a separate acceptable use policy introduces potential restrictions that complicate the notion of ‘open source,’ raising questions about how freely the model can be used and what legal or ethical boundaries might apply.
Yahboom Raspberry Pi 5 ROS2 Robot Car 360°Movement, AI Vision & Tracking, Integrated Multimodal Large AI Model OpenRouter, AI Voice Interaction (Superior Without RPi5)
【Powerful control system】RaspberryPi 5 has made breakthroughs in processor speed,multimedia performance,memory and connection.Based on the RaspberryPi 5 main…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and the Significance of Open Weights
Traditionally, large foundation models have been distributed as closed or semi-closed systems, with access limited through APIs or proprietary licenses. While some models are open source, they often lack transparency in training data and policies. Thinking Machines’ approach—releasing full weights under an open license while maintaining a separate use policy—is unusual and signals a potential shift in how AI models are shared and controlled. The company’s candidness about not claiming to have the strongest model and its emphasis on openness contrast with many industry players, highlighting ongoing debates about transparency, safety, and commercial control in AI development.
“We believe in transparency and giving users the freedom to own and modify their models.”
— Thinking Machines spokesperson
Mastering Small Language Models: A Practical Guide to Building Lightweight NLP Systems with Python, Transformers, and Quantization Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unclear Aspects of Licensing and Use Restrictions
It remains unclear how the separate Model Acceptable Use Policy will be enforced and how it might limit the practical use of Inkling. The policy reportedly prohibits surveillance, deception, and automated decision-making affecting individuals, but the exact scope and legal enforceability of these restrictions are not verified. Additionally, the full training data and pipeline have not been published, leaving questions about transparency in the training process and data sources.
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.
Next Steps for Industry Adoption and Testing
Expect independent researchers and organizations to evaluate Inkling’s performance across various benchmarks and real-world applications. Further transparency about the use policy and potential updates will be closely monitored. Additionally, other AI developers may follow suit, releasing open weights with layered restrictions, potentially reshaping norms around model openness and control. The community will also scrutinize the model’s safety, bias, and ethical implications as adoption grows.
Domain-Specific Small Language Models: Efficient AI for local 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 foundation models?
Inkling is openly available under the Apache 2.0 license, allowing free download, modification, and deployment. It is also multimodal, supporting text, images, and audio, and was trained on a diverse dataset including video and audio.
Does open weight mean the model is fully open source?
No. While the weights are openly available, the training data, pipeline, and any associated policies may be restricted or layered with additional use restrictions, which are not necessarily open source.
Why does the separate use policy matter?
The use policy could impose restrictions on surveillance, deception, and automated decision-making, which may limit how the model can be used despite its open weights. Its enforceability and scope are still uncertain.
Will Inkling outperform proprietary models?
It performs well on several benchmarks, especially in speech and safety, but scores mid-pack or behind some closed models on certain language understanding tasks. Its real-world performance will become clearer with further testing.
What are the implications for AI development?
This release signals a potential shift toward more open, controllable AI models, encouraging transparency and ownership while highlighting ongoing debates about restrictions and safety policies.
Source: ThorstenMeyerAI.com
