📊 Full opportunity report: The Key Insights From Thinking Machines’ Inkling In AI Research on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines released Inkling, a 975-billion-parameter open-weight model under Apache 2.0, marking a shift towards transparency and ownership in AI. Its performance and licensing raise important questions for the industry.
Thinking Machines has released Inkling, a 975-billion-parameter, open-weight foundation model, under the Apache 2.0 license, making it freely downloadable and modifiable. This move marks a significant development in AI transparency and ownership, especially given the model’s open weights and detailed performance disclosures.
Inkling is a multimodal transformer trained on 45 trillion tokens, supporting text, images, and audio inputs with a 1-million-token context window. It features a Mixture-of-Experts architecture with 66 layers and 41 billion active parameters, routing each token to multiple experts for efficiency.
Released openly on Hugging Face under Apache 2.0, the model’s weights are available for download, modification, and deployment, enabling organizations to own and operate their own instances. However, the training data and full pipeline remain proprietary, and the company has reportedly implemented a separate Model Acceptable Use Policy (AUP) restricting certain applications, such as surveillance and automated decision-making.
Thinking Machines also previewed Inkling-Small, a 276-billion-parameter version that reportedly matches or exceeds the larger model’s performance on several benchmarks, with full weights expected after testing. The training process involved hybrid optimization and reinforcement learning, including synthetic data generated by other open models like Kimi K2.5.
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-Weight Release for AI Ownership
The release of Inkling under an open license represents a shift toward greater transparency and control over foundational AI models. Organizations can now download, fine-tune, and deploy the model independently, reducing reliance on proprietary APIs and fostering innovation. However, the presence of a separate AUP raises questions about permissible use, especially in sensitive domains, and highlights ongoing industry debates over true openness versus restricted licensing.
This development could accelerate AI democratization but also complicate regulatory and ethical oversight, as the boundaries of permissible use are less clear when models are openly available but subject to additional restrictions.
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.
Background on Open-Weight Model Releases and Industry Norms
Until now, most large foundation models were released as closed APIs or with restricted weights, limiting direct ownership and modification. While some open models exist, few have approached the scale and multimodal capabilities of Inkling. The industry has largely prioritized proprietary control, with open releases often accompanied by licensing restrictions or limited transparency about training data and processes.
Thinking Machines’ approach, emphasizing open weights with a permissive license, marks a notable departure. The company’s transparency about model specs and performance metrics aligns with growing calls for openness but also introduces new questions about responsible use and enforceability of restrictions.
“Our goal is to empower organizations with ownership and transparency, while maintaining responsible use through our policies.”
— Thinking Machines spokesperson
Multimodal AI Engineering: Practical Systems and Tools for Developers and AI Engineers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Inkling’s Use Restrictions
It remains unclear how the separate Model Acceptable Use Policy (AUP) will be enforced and whether it will impact the ability of organizations to freely modify and deploy Inkling. The exact scope of restrictions and how they compare to the permissiveness of Apache 2.0 licensing are still to be verified. Additionally, the full training data and pipeline have not been disclosed, raising questions about transparency and reproducibility.
Kimi K3 for Owners: The Grounded Guide to the 2.8-Trillion-Parameter Open Model: What It Really Does, What the Hype Gets Wrong, and Why Owning Your Context Beats Chasing the Model
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Industry Adoption and Policy Clarification
Expect further testing and independent benchmarking of Inkling’s performance, alongside scrutiny of the AUP’s terms and enforceability. Organizations interested in adopting the model will likely evaluate the legal and ethical implications of the layered restrictions. Meanwhile, the industry will watch for additional open releases and clarifications from Thinking Machines regarding data transparency and use policies.
Laplink PCmover – Easy Migration of your Applications, Files and Settings from an Old PC to a New PC – Data Transfer Software with Optional Super Speed USB 3.0 Cable – Business Standard, 10 Licenses
- Flexible Licensing Tiers: 1, 5, 10, or 25 transfers
- Unlimited Transfer Attempts: One license per source-destination pair
- Suitable for Multiple PCs: Supports individual or multi-PC migrations
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What makes Inkling different from other foundation models?
Inkling is notable for its open weights released under Apache 2.0, supporting ownership and customization, along with its multimodal input capabilities and large parameter count (975B). It also emphasizes transparency about performance metrics.
Does open weights mean full open source?
No. While the weights are openly available, the training data, pipeline, and potentially some use restrictions are not disclosed. Additionally, a separate AUP may impose restrictions beyond the license.
What are the potential risks of using Inkling?
Risks include unclear enforcement of use restrictions, potential ethical concerns related to the AUP, and the lack of transparency about training data, which could impact bias and safety assessments.
How might this influence future AI model releases?
This could encourage more open models with clear licensing, but also lead to more layered restrictions and policies to manage responsible use, shaping the industry’s approach to openness and control.
Source: ThorstenMeyerAI.com
