📊 Full opportunity report: The Weights Came First: What Thinking Machines’ Inkling Actually Signals on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
Open a free Amazon Business account
Business pricing, bulk buying and tax-exempt orders.
Create a free accountAs an affiliate, we earn on qualifying purchases.
TL;DR
Thinking Machines publicly released the full weights of its new multimodal model, Inkling, under Apache 2.0 license, emphasizing transparency over claiming dominance. The move signals a shift towards open access, but questions remain about licensing restrictions and data transparency.
Thinking Machines has publicly released the full weights of its Inkling model on Hugging Face, making it available under the Apache 2.0 license. This move is significant because it prioritizes transparency and ownership over dominance, contrasting with typical industry practices of withholding weights or releasing closed APIs. The release signals a shift in how foundational models are distributed and used, especially amid ongoing debates over open access and licensing in AI development.
The Inkling model is a 975-billion-parameter mixture-of-experts transformer supporting multimodal input—text, images, and audio—processed jointly without an external vision adapter. It was pretrained on 45 trillion tokens across various media types and supports a 1-million-token context window. The full weights are now available on Hugging Face under Apache 2.0, allowing users to download, modify, and deploy independently. This is a departure from the norm, where companies often release models with closed licenses or only provide API access.
While the weights are openly accessible, the company has reportedly implemented a separate Model Acceptable Use Policy restricting surveillance, deception, and automated decision-making affecting individuals’ rights. This layered policy introduces questions about the true openness of the model, as the Apache license itself imposes no such restrictions. Additionally, the training data and full pipeline have not been publicly disclosed, which is typical but limits full transparency. The model’s performance claims include strong results in speech and safety benchmarks but middling scores in pure language understanding tasks.
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 Weights and Licensing Restrictions
This move by Thinking Machines underscores a growing trend toward open access to foundational AI models as a way to foster innovation, transparency, and ownership. Releasing the full weights under an open license allows organizations to fine-tune, deploy, and inspect the model independently, reducing reliance on API-based access. However, the presence of a separate Acceptable Use Policy layered on top of the open license raises questions about the true extent of openness and control. For industries concerned with data privacy, security, and compliance—such as public safety or geospatial applications—these restrictions could influence adoption decisions. Overall, this signals a potential shift in industry norms, emphasizing ownership and transparency over proprietary control.
LOCAL LLM DEPLOYMENT: Training, Fine-Tuning, & Offline Inference: The Complete Developer’s Guide to Building, Training, and Running Private Open-Source AI Offline (with full source code)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and the Significance of Open Model Releases
Historically, most large AI models have been released with closed licenses or API-only access, limiting user control and inspection. Recent high-profile incidents, such as government directives to disable models, have heightened awareness around the importance of owning and controlling AI assets. The release of Inkling’s weights under Apache 2.0 marks a notable departure, aligning with broader calls for open-source foundations in AI. Nonetheless, most companies still withhold full training data and pipelines, citing proprietary concerns, which complicates claims of full transparency. The industry remains divided over the balance between openness and control, with recent launches signaling a possible shift towards more open models, especially in safety-critical sectors.
Unresolved Questions About Licensing and Data Transparency
It remains unclear how the separate Acceptable Use Policy will be enforced and whether it will effectively limit the model’s open use. The full training data, pipeline details, and evaluation benchmarks beyond select claims have not been disclosed, raising questions about full transparency and reproducibility. Additionally, the impact of layered restrictions on commercial or research use has yet to be tested in practice. The extent to which users can freely modify and deploy Inkling without restrictions remains an open question, as does the potential for future policy changes.
Next Steps for Adoption and Policy Clarification
Expect further clarification from Thinking Machines regarding the scope and enforcement of its use policy. Industry observers will monitor whether other organizations follow suit by releasing open weights with layered restrictions or layered policies. Independent researchers and competitors will likely attempt to replicate the benchmarks and verify claims, especially around safety and multimodal performance. Additionally, users and developers will test the practical limits of ownership—modifying, fine-tuning, and deploying Inkling across various applications—while awaiting full transparency on training data and pipelines.
Key Questions
What makes Inkling different from other foundational models?
Inkling is a multimodal, mixture-of-experts transformer with 975 billion parameters, supporting text, images, and audio inputs, pretrained on 45 trillion tokens, and released with open weights under Apache 2.0 license.
What restrictions are associated with Inkling’s open weights?
While the weights are openly available, reports suggest a separate Acceptable Use Policy restricts surveillance, deception, and automated decision-making, which may limit practical openness.
Why is the layered use policy significant?
The layered policy introduces legal and practical restrictions beyond the open license, raising questions about the true level of openness and control over the model’s use.
Will the full training data be released?
No, the training data and full pipeline details have not been publicly disclosed, which limits complete transparency and reproducibility.
What does this mean for the future of open AI models?
This move suggests a potential shift toward more open models with layered restrictions, balancing transparency with responsible use, especially in safety-critical sectors.
Source: ThorstenMeyerAI.com
Back to school Picks
back to school
As an affiliate, we earn on qualifying purchases.
