📊 Full opportunity report: The Weights Came First: What Thinking Machines’ Inkling Actually Signals on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines publicly released the full weights of its Inkling model under an open license, emphasizing transparency. This move challenges industry norms by prioritizing open access over claiming top performance. The release raises questions about licensing, use policies, and the model’s actual capabilities.
Thinking Machines has released the full weights of its Inkling model on Hugging Face under an Apache 2.0 license, making it openly available for download, modification, and deployment. This is notable because most large foundation models are either closed or released with restrictive licenses, but Inkling’s release emphasizes transparency and ownership.
The Inkling model is a 975-billion-parameter Mixture-of-Experts transformer supporting multimodal inputs—text, images, and audio—trained on 45 trillion tokens. Its weights are now accessible under an open license, allowing organizations to fine-tune and deploy independently. The release includes a smaller variant, Inkling-Small, with 276 billion parameters, which reportedly matches or exceeds performance benchmarks of larger models.
Unlike typical industry releases, the model’s weights are available openly, but the training data and full training pipeline are not published. Additionally, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP) that restricts certain applications, such as surveillance and deception, which could complicate the open-source nature of the release. The company stated that the weights are not the strongest available, but the transparency is intentional—aimed at fostering open research and ownership.
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 for Industry and Users
This release marks a shift toward greater transparency in the AI industry, allowing organizations to own and modify models rather than rent or license them. It challenges the norm of proprietary, closed models and could influence future open-source practices. However, the existence of a separate AUP raises questions about the scope of openness and potential restrictions on use, which could impact trust and adoption among certain sectors.
![INIU USB C to USB C Cable, 240W Fast Charging Type C Charger Cord, Braided USBC Phone Charger Cable for iPhone 17 16 Pro Max Samsung S26 S25 Laptops MacBook Air 15 M3 iPad Switch [6.6ft+6.6ft Black]](https://m.media-amazon.com/images/I/41r3DXOHFvL._SL500_.jpg)
INIU USB C to USB C Cable, 240W Fast Charging Type C Charger Cord, Braided USBC Phone Charger Cable for iPhone 17 16 Pro Max Samsung S26 S25 Laptops MacBook Air 15 M3 iPad Switch [6.6ft+6.6ft Black]
[From INIU—the SAFE Fast Charge Pro] Experience the safest charging with over 38 million global users. At INIU,…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and Recent Model Releases
Most large foundation models, including those from leading AI labs, are released with restrictions or as closed models, limiting access to weights and training data. Recently, some companies have begun releasing models with open weights, but often with caveats or restrictive licenses. The release of Inkling’s weights under Apache 2.0 is unusual in that it emphasizes openness while simultaneously hinting at usage restrictions through a separate policy. This approach reflects ongoing tensions between transparency, commercial interests, and responsible AI deployment.
“Our goal is to foster open research and give users control over the models they deploy.”
— Thinking Machines spokesperson
The Hardware Hacking Handbook: Breaking Embedded Security with Hardware Attacks
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Open Source Status and Usage Restrictions Clarified
It remains unclear whether the separate Model Acceptable Use Policy (AUP) imposes restrictions that limit the true openness of the release. The full scope, enforceability, and implications of this policy are not yet verified, raising questions about how open the model truly is in practice.
Hisense 43" E6 Cinema Series Hi-QLED 4K UHD Smart Fire TV (43E6QF) – AI Light Sensor, Dolby Vision · Atmos, Voice Remote with Alexa, Motion Rate 120, HDR 10+ Adaptive, Game Mode Plus
HI-QLED COLOR: Experience a visual treat with Hi-QLED Color. Every frame is rich, vibrant, and true-to-life, ensuring that…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps in Model Adoption and Policy Clarification
Organizations and researchers will likely test and benchmark Inkling’s performance and assess the legal and ethical restrictions imposed by the AUP. Further disclosures from Thinking Machines regarding the full training pipeline, data sources, and policy details are expected, which will clarify the model’s openness and usability.
Accelerate Everything with Tensor Cores: A Developer’s Guide to High-Performance AI, Efficient Training, and Scalable Models
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 significant?
The open release allows organizations to own, modify, and deploy the model independently, challenging industry norms of proprietary models and promoting transparency.
Does the open weights mean the model is fully open source?
Not necessarily. While the weights are under Apache 2.0, reports suggest a separate Acceptable Use Policy may impose restrictions, which complicates the model’s openness.
What are the potential risks of this open release?
If restrictions exist through the AUP, misuse or unintended applications could be limited despite the open weights. Additionally, lack of transparency about training data raises concerns about bias and safety.
How does this compare to other recent model releases?
Most recent releases have been either closed or with restrictive licenses; Inkling’s open weights under Apache 2.0 mark a notable departure toward transparency and user ownership.
Source: ThorstenMeyerAI.com
