📊 Full opportunity report: Understanding The Implications Of Thinking Machines’ Inkling In AI on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released its new AI model, Inkling, under an open license, but clarifies it is not the strongest available. The release emphasizes transparency and ownership benefits, raising questions about licensing and use restrictions.
Thinking Machines has officially released the full weights of its new AI model, Inkling, under an open-source license, making it accessible for download and modification. This move marks a notable shift in the AI landscape, emphasizing transparency and ownership over proprietary control, and directly addresses the ongoing debate about the costs and benefits of owning versus renting AI models.
Inkling is a 975-billion-parameter Mixture-of-Experts transformer supporting multimodal input—text, images, and audio—trained on 45 trillion tokens across various media. It features a 66-layer decoder-only architecture with a 1-million-token context window. The full weights were published openly on Hugging Face under the Apache 2.0 license, allowing users to download, modify, and deploy independently.
Despite its open release, the company clarified that Inkling is not the strongest model available, with benchmarks showing it performs well in safety and speech but less so in some language understanding tasks. The release also included a smaller variant, Inkling-Small, which matches or exceeds the larger model on several benchmarks. The company disclosed details of the training process, including the use of synthetic data from open models like Kimi K2.5, and a hybrid optimizer approach.
However, questions remain regarding the licensing restrictions. Reports suggest that Thinking Machines imposes a separate Model Acceptable Use Policy that may limit surveillance, deception, and automated decision-making, potentially conflicting with the open-source license. The company has not published the full policy, leaving uncertainty about usage constraints.
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-Source Release and Usage Restrictions
The release of Inkling under an open license represents a significant step toward democratizing access to large-scale AI models, enabling wider experimentation and deployment outside proprietary ecosystems. However, the potential layering of usage restrictions through a separate policy complicates the narrative of true openness. This development could influence how organizations consider owning or licensing AI models, especially in sensitive domains like public safety or surveillance, where restrictions could impact deployment and compliance.
The MCP Standard: A Developer's Guide to Building Universal AI Tools with the Model Context Protocol
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Recent Trends in Open AI Model Releases and Industry Norms
Over the past year, several AI labs have moved toward open-sourcing models or releasing weights openly, driven by community demands for transparency and control. Notably, Meta, Stanford, and others have released models under permissive licenses, contrasting with proprietary approaches from companies like OpenAI. Thinking Machines’ approach with Inkling—full weights released openly but with an additional usage policy—fits into this evolving landscape, highlighting tensions between openness, control, and safety concerns.
Historically, open releases have often been accompanied by restrictions or unclear licensing terms, leading to debates about true openness. Inkling’s release underscores these ongoing tensions, especially as models grow larger and more capable.
“Our goal is to foster transparency and ownership, giving users full control over the model while maintaining responsible use policies.”
— A representative from Thinking Machines
Multimodal AI Systems Engineering: Building Production Vision-Language Models, Document AI, and Cross-Modal Retrieval Pipelines (Production AI Engineering Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unclear Aspects of Licensing and Usage Policies
It remains unclear whether the separate Model Acceptable Use Policy explicitly restricts certain applications or if it is enforceable against all users. The full text of the policy has not been made public, and the extent of restrictions—if any—beyond the license itself is uncertain. Additionally, how this layered approach will influence legal and commercial use remains to be seen, especially in sensitive sectors.
NVD RTX PRO 6000 Blackwell Professional Workstation Edition Graphics Card for AI, Design, Simulation, Engineering – 96GB DDR7 ECC Memory – 4th Gen RT/5th Gen Tensor Core GPU – OEM Packaging
[NVIDIA Blackwell Streaming Multiprocessor] The new SM features increased processing throughput, and new neural shaders that integrate neural…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Model Adoption and Policy Clarification
Expect further clarification from Thinking Machines regarding the full details of the Acceptable Use Policy and its enforceability. Industry observers will likely scrutinize how organizations implement and comply with the restrictions, especially in regulated domains. Additionally, independent benchmarking and real-world testing will determine Inkling’s competitiveness and safety profile relative to other models.
AI model licensing and restrictions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Is Inkling truly open-source?
Yes, the full weights are released under the Apache 2.0 license, allowing download, modification, and deployment. However, reports suggest a separate usage policy may impose restrictions, which complicates the notion of full openness.
What are the main capabilities of Inkling?
Inkling supports multimodal input—text, images, and audio—with a 1-million-token context window and has demonstrated strong safety performance and speech capabilities. Its language understanding benchmarks are more moderate.
Does the licensing restrict certain applications?
It is not yet confirmed. Reports indicate a separate Model Acceptable Use Policy that may limit surveillance, deception, and automated decision-making, but the full policy has not been publicly released for verification.
How does Inkling compare to other models?
In benchmarks, Inkling performs well in safety and speech but is mid-tier in language understanding tasks. Its open-weight approach allows for independent testing and customization.
What does this mean for AI openness?
This release exemplifies a nuanced approach—full weights are open, but usage restrictions may apply—highlighting ongoing debates about true openness versus controlled deployment in AI development.
Source: ThorstenMeyerAI.com
