📊 Full opportunity report: What The Inkling From Thinking Machines Means For AI Researchers on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has launched Inkling, a large open-weight AI model with 975 billion parameters, available under Apache 2.0. The release emphasizes transparency but raises questions about licensing and use restrictions, impacting AI development and open-source policies.
Thinking Machines has publicly released the full weights of its new AI model, Inkling, under the open-source Apache 2.0 license. This move is notable because the model is not the most powerful available, but it emphasizes transparency and ownership, challenging industry norms.
The Inkling model is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active. It supports a one-million-token context window and was trained on 45 trillion tokens across text, images, audio, and video modalities. The model’s architecture includes a decoder-only design routing tokens to multiple experts, and it was trained using a hybrid optimizer on NVIDIA systems.
Importantly, the full weights are now available on Hugging Face under Apache 2.0, allowing users to download, modify, and deploy the model independently. This is a departure from typical industry practice where models are often released with restrictions or only as API access. However, the release comes with caveats: the training data and pipeline are not published, and there are reports of a separate Acceptable Use Policy that restricts surveillance, deception, and automated decision-making.
Some benchmarks show Inkling performing strongly in safety and speech tasks, but it is mid-pack or behind in certain language understanding tests. The model’s release has sparked debate about the implications of open weights versus open source, especially given the additional use restrictions reportedly layered on top of the Apache license.
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 Weights and Licensing Restrictions
The release of Inkling’s full weights under an open license marks a significant shift toward transparency and user ownership in AI development. It enables organizations to fine-tune, inspect, and deploy the model independently, which could accelerate innovation and reduce reliance on proprietary APIs.
However, the reported presence of a separate, more restrictive Acceptable Use Policy raises questions about the true openness of the model. If such restrictions are enforceable, they could limit the model’s use in sensitive domains like surveillance or decision-making, potentially undermining the benefits of open weights. This tension between open licensing and layered restrictions could influence industry standards and regulatory discussions around AI transparency and control.
AI Systems Performance Engineering: Optimizing Model Training and Inference Workloads with GPUs, CUDA, and PyTorch
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and the Shift Toward Transparency
Historically, most large AI models have been released with limited access, either via API or with restrictions on weights, to control misuse and protect intellectual property. Recent developments, including Meta’s open releases and now Thinking Machines’ Inkling, indicate a growing trend toward releasing full model weights to foster transparency and community-driven innovation.
Nevertheless, the industry remains divided: some companies emphasize open source for collaboration, while others prioritize restricting access to prevent misuse. Inkling’s release under Apache 2.0, combined with the reported use restrictions, exemplifies this ongoing tension, highlighting evolving debates about what true openness entails in AI.
“We believe in transparency and empowering users to own and adapt our models, but we also have policies to prevent misuse.”
— Thinking Machines spokesperson
Unclear Aspects of Licensing and Use Restrictions
It remains unclear how enforceable the reported Acceptable Use Policy (AUP) is, and whether it legally constrains the use of the open weights under Apache 2.0. The exact scope and enforceability of these layered restrictions are still under investigation, and potential users should verify the policies before deploying the model.
Additionally, the impact of the training data’s proprietary nature and the absence of detailed training pipeline documentation add further uncertainty about the model’s transparency and replicability.
Future Steps for Model Adoption and Policy Clarification
Next, AI researchers and organizations will scrutinize the licensing and use policies, testing the enforceability of restrictions. Independent audits and benchmark tests are expected to follow to verify the model’s capabilities and safety claims. Further clarifications from Thinking Machines regarding the AUP and training data will be crucial for broader adoption.
Industry discussions on open-source standards, licensing, and responsible AI use are likely to intensify, shaping future norms and regulations.
Key Questions
What makes Inkling different from other large language models?
Inkling is notable for being released with full weights under the Apache 2.0 license, allowing unrestricted download, modification, and deployment. It also supports multimodal input (text, images, audio) and has a large context window of one million tokens.
Are there restrictions on how I can use Inkling?
While the weights are openly available, reports suggest there is a separate Acceptable Use Policy that restricts surveillance, deception, and automated decision-making affecting individuals. The enforceability of these restrictions is still uncertain.
Why is the licensing approach significant?
Releasing full weights under Apache 2.0 promotes transparency and ownership, enabling independent fine-tuning and deployment. However, layered restrictions complicate the notion of true open source and may influence industry standards.
What are the potential risks of open weights with restrictions?
Restrictions could limit the model’s use in sensitive areas, and if enforceable, might undermine the benefits of open access, raising questions about the true openness and safety of the model.
What are the next steps for AI developers regarding Inkling?
Developers will evaluate the enforceability of restrictions, test the model’s capabilities, and seek clarifications from Thinking Machines. The broader industry will also debate standards for open-source AI models.
Source: ThorstenMeyerAI.com
