📊 Full opportunity report: The Weights Came First: Unpacking AI’s Hidden Messages on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines publicly released the full weights of its Inkling AI model on Hugging Face under Apache 2.0 license, emphasizing transparency. This contrasts with typical industry practices of closed models and raises questions about open source claims and usage policies.
Thinking Machines has released the full weights of its new Inkling AI model on Hugging Face under an open-source license, marking a significant departure from industry norms. This move makes the weights freely available for download, modification, and deployment, even as the company states that Inkling is not the strongest model available. The release is notable because most companies typically restrict access through APIs or closed licenses, rather than releasing weights openly at launch.
The Inkling model is a 975-billion parameter mixture-of-experts transformer supporting a 1-million-token context window. It was trained on 45 trillion tokens across text, images, audio, and video, with a multimodal input design that processes text, images, and audio natively without vision adapters. The full weights were uploaded to Hugging Face under Apache 2.0 license, enabling anyone to download, fine-tune, and deploy the model independently.
Unlike many recent AI launches, which restrict access via APIs or proprietary licenses, Thinking Machines opted for transparency by providing open weights first, with day-zero support in popular frameworks like transformers, vLLM, SGLang, and llama.cpp. The company also disclosed that the training involved hybrid optimization and synthetic data generated by other open-weight models, including the Chinese Kimi K2.5 model. However, the company also reportedly maintains a separate Model Acceptable Use Policy (AUP) that restricts surveillance, deception, and fully automated decisions affecting individuals’ rights, raising questions about the true openness of the release.
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-First Model Release in AI Industry
This release challenges the industry norm of proprietary, API-only access to powerful AI models, emphasizing transparency and user control. It enables organizations to independently inspect, modify, and deploy Inkling, potentially accelerating innovation and reducing reliance on closed systems. However, the existence of a separate AUP suggests that the notion of ‘true open source’ may be more nuanced, raising questions about how open the model truly is and what restrictions apply in practice. The move could influence future model releases, prompting more companies to consider open weights as a strategic choice.
Fine-Tuning Large Language Models: From Custom Datasets to High-Performance AI Models Using Modern Toolchains
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Shift Toward Open Model Weights
Historically, most AI companies have released models through APIs or closed licenses, citing concerns over misuse, safety, and commercial interests. The recent trend has been toward proprietary models with restricted access, even when models are technically open in terms of licensing. Thinking Machines’ decision to release Inkling’s weights openly on day one marks a notable shift, aligning with broader calls for transparency and democratization in AI development. Prior to this, only a few organizations, such as Stability AI, had released open weights at similar stages, but Inkling’s approach is among the most comprehensive to date.
This move comes amid ongoing debates about the balance between openness, safety, and commercial viability, with some experts arguing that open models require robust governance frameworks to prevent misuse. The release also follows recent incidents where proprietary models were turned off or restricted, highlighting the potential benefits of open weights for resilience and independence.
“Our goal is to promote transparency and innovation by making Inkling’s weights openly available, while maintaining responsible use policies.”
— Thinking Machines spokesperson
From Weights to Wisdom: The Complete Guide to Running and Adapting Opensource AI Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Clarifications Needed on Usage Restrictions and Open Source Claims
It is still unclear how the separate Model Acceptable Use Policy (AUP) will be enforced and whether it significantly limits the open-source nature of the weights. The AUP reportedly prohibits surveillance, deception, and automated decision-making affecting individuals, but the specifics and legal enforceability remain unverified. Additionally, the full training data and pipeline have not been released, which raises questions about the true extent of transparency and openness.
HIWONDER Humanoid Robot with ChatGPT Multimodal AI Models AI Embodied Intelligent Vision Scene Voice Understanding 20DOF Educational Robot Kit Python Programming, TonyPi Advanced & RaspberryPi 5 4GB
- Powered by Raspberry Pi 5: High-performance AI vision robot
- Open-source development platform: Supports advanced AI robotics development
- ChatGPT multimodal integration: Enhanced human-machine interaction
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Industry Adoption and Policy Development
Expect other AI developers to observe and potentially follow suit if Inkling’s open weights demonstrate practical advantages. Further discussions are likely on establishing standardized governance for open models, especially regarding safety and misuse prevention. Companies and researchers will scrutinize the AUP and licensing terms, and independent testing will assess the model’s capabilities and safety features in real-world applications. The ongoing debate about open versus restricted models will intensify as more organizations weigh the benefits of transparency against risks.
Fine-Tuning Large Language Models: From Custom Datasets to High-Performance AI Models Using Modern Toolchains
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is the open release of Inkling’s weights significant?
The open release allows organizations and developers to freely access, modify, and deploy the model, promoting transparency and innovation in AI development, contrasting with typical proprietary approaches.
Does releasing weights mean the model is fully open source?
Not necessarily. While the weights are under Apache 2.0 license, the training data, pipeline, and potential usage restrictions via the AUP are not fully disclosed, complicating the open source claim.
What concerns exist around the separate usage policy?
If the AUP restricts certain uses, it could limit the practical openness of the model, raising questions about how freely it can be used despite the open weights license.
How might this impact the AI industry?
This move could encourage more open releases, pushing the industry toward greater transparency, but also sparks debate over safety, misuse, and governance of open models.
Source: ThorstenMeyerAI.com
