📊 Full opportunity report: AI’s Hidden Messages: Insights From Thinking Machines’ First Inkling on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines released Inkling, a 975-billion-parameter multimodal AI model, openly on Hugging Face under Apache 2.0. While openly available, it faces restrictions via a separate use policy, raising questions about true openness.
Thinking Machines has publicly released its first foundation model, Inkling, under an open license on Hugging Face, marking a significant step in transparency for large AI models. This move is notable because the company explicitly stated that Inkling is not the strongest model available, emphasizing honesty about its capabilities and limitations.
Inkling is a 975-billion-parameter, multimodal transformer supporting a 1-million-token context window. It was pretrained on 45 trillion tokens, including text, images, audio, and video, and features a shared multimodal processing architecture that integrates audio spectrograms and image patches directly into its shared space. The model was trained using a hybrid optimizer on NVIDIA systems, with over 30 million reinforcement learning rollouts that improved reasoning performance.
Released openly on Hugging Face under Apache 2.0 license, Inkling allows users to download, modify, and deploy the model independently. However, there are important caveats: the training data and full training pipeline are not public, and reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy that restricts surveillance, deception, and automated decision-making impacting individuals. These restrictions could limit the model’s true open-source status.
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 Release and Usage Restrictions
The release of Inkling under an open license with restrictions highlights ongoing tensions in AI transparency. While the model’s weights are freely available, the potential for usage limitations through a separate policy raises questions about true openness. This development could influence how organizations approach open-source AI, balancing transparency with ethical and legal constraints, especially in sensitive domains like security or public safety.
AI development and training kits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Large-Scale Open AI Models
In recent months, the AI community has seen a surge in large models being released with open licenses, aiming to foster transparency and democratize access. However, most releases have been accompanied by restrictions or unclear licensing terms. Thinking Machines’ approach with Inkling—especially its candid acknowledgment that it is not the most capable model—marks a shift towards more honest communication about model strengths and limitations, contrasting with some competitors’ more boastful claims.
The company’s decision to publish the full weights first, along with detailed training methodology, is a notable move in the ongoing debate over open-source AI. It follows recent incidents where models were switched off or restricted by authorities, emphasizing the importance of control and independent deployment.
“Our goal is to promote open innovation while respecting ethical boundaries and user safety.”
— Mira Murati, Thinking Machines CEO
multimodal AI model hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unclear Aspects of Inkling’s Usage Restrictions
Details about the Model Acceptable Use Policy remain unverified, and it is not confirmed how strictly the restrictions will be enforced or how they might impact independent developers and organizations. The extent to which these restrictions could limit practical use or modify the open-source nature of Inkling is still uncertain.
AI model licensing and deployment books
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Independent Testing
Independent researchers and organizations are expected to evaluate Inkling’s performance, safety, and compliance with its use policy. Further disclosures from Thinking Machines about the policy’s specifics and real-world enforcement will clarify its openness. Additionally, benchmarks and real-world applications will determine how widely the model is adopted and trusted.
AI model training data management tools
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 model weights are released under Apache 2.0 license, reports suggest there may be additional restrictions via a separate use policy. The full openness depends on how those restrictions are enforced.
What makes Inkling different from other large models?
Inkling is a multimodal, 975-billion-parameter transformer with a focus on transparency, openly available weights, and detailed training methodology, but with potential restrictions on usage.
Why does the licensing matter?
The license determines whether users can freely modify, deploy, and commercialize the model. Restrictions through additional policies can limit these freedoms despite an open license.
What are the ethical concerns around open models?
Open models can be misused for surveillance, deception, or harmful automation. Restrictions aim to mitigate these risks but may limit legitimate uses.
What is the significance of the training data details?
The training data and pipeline are not publicly disclosed, which raises questions about transparency and reproducibility despite open weights.
Source: ThorstenMeyerAI.com
