📊 Full opportunity report: The Hidden Meaning Behind Thinking Machines’ Inkling In AI Progress on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large multimodal AI model, under an open license but with restrictions via a separate policy. This raises questions about true openness and control.
Thinking Machines has publicly released its first foundation model, Inkling, under an Apache 2.0 license. This marks a significant step in AI model distribution, emphasizing open access to weights, yet it also introduces restrictions through a separate Model Acceptable Use Policy, complicating the notion of full open-source availability.
Inkling is a 975-billion-parameter multimodal transformer supporting text, images, and audio inputs. It was trained on 45 trillion tokens across multiple modalities and supports a 1-million-token context window. The model’s weights were released openly on Hugging Face, enabling organizations to fine-tune, deploy, and modify the model independently.
However, the release is accompanied by a separate Acceptable Use Policy that restricts certain applications, such as surveillance, deception, and automated decision-making affecting individuals’ rights. This policy raises questions about the true openness of the model, as the Apache 2.0 license itself imposes no such restrictions.
Further, the model’s training data and full training pipeline remain unpublished, which is typical but limits full transparency. The model’s performance on various benchmarks shows strengths in safety and speech recognition, but it is mid-pack on some language understanding tests, reflecting a balanced but not state-of-the-art performance profile.
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 Restricted Open Licensing for AI Models
This release highlights the ongoing debate over what constitutes truly open-source AI models. While Inkling’s weights are freely available under Apache 2.0, the existence of a separate use policy that limits certain applications suggests that ‘open’ may not mean entirely unrestricted. For organizations and developers, this complicates licensing decisions, especially in sensitive domains like public safety or surveillance.
Moreover, the release demonstrates a shift toward transparency about model capabilities and limitations, even if restrictions are layered on top. It signals a move toward more responsible AI deployment, but also raises concerns about the potential for restrictions to be enforced selectively or ambiguously.
multimodal AI development tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on AI Model Releases and Openness
In recent years, the AI community has grappled with the tension between open model sharing and safeguarding against misuse. Major players like OpenAI and Meta have released models with varying degrees of openness, often accompanied by usage restrictions or licensing conditions.
Thinking Machines, founded by former OpenAI CTO, has taken a different approach by releasing weights openly on Hugging Face, aligning with a trend toward democratizing access. However, the layered restrictions via a separate policy mark a nuanced evolution in the open-source paradigm, reflecting industry concerns about safety, misuse, and control.
This development follows recent instances where governments and organizations have restricted access to powerful models, emphasizing the importance of balancing openness with responsible use.
“We believe in open access to our models, but responsible use is our priority. The separate policy ensures that our technology is used ethically.”
— Thinking Machines spokesperson
AI model fine-tuning software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Inkling’s Open-Source Status
It remains unclear how enforceable the separate Model Acceptable Use Policy is in practice and whether it effectively limits the model’s use in certain domains. The full training data and pipeline are not published, which limits transparency about the model’s origins and biases. Additionally, the long-term impact of layered restrictions on open-source AI development is still uncertain.
large language model training kits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Evaluating Inkling’s Adoption and Impact
Organizations and developers will likely scrutinize the licensing and use policy before adopting Inkling for sensitive applications. Further independent benchmarking and transparency about the training data and policy enforcement are expected. Additionally, other AI labs may follow suit, balancing openness with restrictions, shaping future standards in AI model sharing.
Local LLM Inference Optimization: A Comprehensive Guide to Quantization, Hardware Acceleration, and Efficient Private AI Deployment
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Is Inkling truly open source?
While the weights are released under Apache 2.0, a separate Use Policy imposes restrictions, complicating the notion of full open-source status.
What restrictions are included in the use policy?
The policy reportedly prohibits surveillance, deception, and fully automated decision-making affecting individuals’ rights, though the enforceability remains unclear.
How does this release compare to previous open models?
Unlike many open models, Inkling’s release includes layered restrictions, making it a hybrid approach that combines open access with responsible use considerations.
What are the implications for AI safety and ethics?
The layered restrictions aim to promote ethical use, but they also introduce ambiguity about what is permissible, raising ongoing debates about openness versus control.
Will this influence future AI model releases?
Likely yes. The approach signals a trend toward more transparent licensing combined with responsible use policies, shaping industry standards.
Source: ThorstenMeyerAI.com
