📊 Full opportunity report: Could Thinking Machines’ Inkling Predict AI’s Future? on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large open-access AI model, openly stating it is not the top performer. This move highlights the shift toward transparency and ownership in AI development.
Thinking Machines has released its first foundation model, Inkling, making it available in full on Hugging Face under the Apache 2.0 license. The company explicitly states that Inkling is not the strongest model available, marking a notable shift toward transparency about AI model performance and ownership.
Inkling is a Mixture-of-Experts transformer with 975 billion parameters and a 66-layer decoder-only architecture supporting a 1-million-token context window. It was pretrained on 45 trillion tokens, including text, images, audio, and video, and is natively multimodal, processing inputs from text, images, and audio without additional vision adapters. The full weights are now publicly available on Hugging Face, with the model licensed under Apache 2.0, allowing download, modification, and commercial use.
Thinking Machines emphasizes transparency: the weights are open, but the training data and pipeline are not published. The company also reportedly maintains a separate Model Acceptable Use Policy restricting surveillance, deception, and automated decision-making affecting individuals, which introduces a layer of restrictions beyond the open license. The model’s performance claims include strong results in safety benchmarks and speech recognition, but it ranks mid-tier on some language understanding benchmarks.
This release is significant because it prioritizes model ownership and transparency, contrasting with many proprietary models that are only accessible via API. It also raises questions about the scope and enforceability of the company’s use policies, which could influence how organizations adopt open models in sensitive domains.
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-Access Model Release
The release of Inkling under an open license with full weights represents a shift toward ownership and transparency in AI development. It allows organizations to fine-tune, inspect, and deploy the model independently, reducing reliance on API-based access. However, the reported Model Acceptable Use Policy suggests restrictions that could complicate deployment in areas like surveillance or automated decision-making, potentially limiting the model’s practical applications in sensitive sectors. This move may influence future industry standards for open models and ownership rights, encouraging more transparency but also raising questions about control and misuse.
NVD RTX PRO 6000 Blackwell Professional Workstation Edition Graphics Card for AI, Design, Simulation, Engineering – 96GB DDR7 ECC Memory – 4th Gen RT/5th Gen Tensor Core GPU – OEM Packaging
- Enhanced Processing Power: NVIDIA Blackwell Streaming Multiprocessor with neural shaders
- Advanced Cooling Design: Double-flow-through cooling for peak performance
- Next-Gen Tensor Cores: 5th Gen Tensor Cores deliver 3X performance
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Open-Weight Model Releases
In recent months, the AI community has seen a growing push for open models, driven by concerns over transparency, control, and commercial ownership. Unlike proprietary models from companies like OpenAI or Google, open-weight models enable users to inspect, modify, and deploy AI systems independently. Thinking Machines, founded by former OpenAI CTO, has positioned itself as a challenger by releasing Inkling with full weights publicly available, emphasizing honesty about its performance and limitations. This approach contrasts with earlier models that often kept weights proprietary or restricted access, fostering debates over openness versus safety and misuse risks.
The trend reflects broader industry discussions about the balance between openness and responsible use, especially after incidents where models were shut down or restricted due to policy or regulatory concerns. Inkling’s release is a notable milestone in this evolving landscape.
“We believe in providing full access to our models while maintaining responsible use policies. Transparency is key to advancing AI safely.”
— Thinking Machines spokesperson
Unresolved Questions About Inkling’s Use Policies
It is not yet clear how the reported Model Acceptable Use Policy will be enforced or how it will impact practical deployment, especially in sensitive domains. The details of the restrictions and their scope remain unverified, and there is ongoing debate about whether such layered policies conflict with the open-source license. Additionally, the actual performance of Inkling in diverse real-world applications needs further independent validation, as current benchmarks are vendor-reported and not yet fully peer-reviewed.
Next Steps for Inkling’s Adoption and Evaluation
Independent researchers and organizations will likely conduct further benchmarking and testing of Inkling to verify performance claims. The company may release additional details about its use policies and training data. Adoption in commercial and sensitive sectors will depend on how the restrictions are interpreted and enforced. Future updates may include more detailed safety evaluations, real-world case studies, and potential community-driven improvements or modifications.
Key Questions
What makes Inkling different from other foundation models?
Inkling is openly available with full weights under the Apache 2.0 license, allowing users to download, modify, and deploy it independently. It emphasizes transparency about its performance and limitations, unlike many proprietary models.
Does open access mean the model is completely unrestricted?
No. While the weights are open, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy that restricts certain applications, such as surveillance and automated decision-making affecting individuals.
How does Inkling perform compared to other models?
In benchmarks, Inkling scores highly on safety and speech recognition tasks but is mid-tier on some language understanding benchmarks. Its performance is openly acknowledged as not the top among current models.
What are the risks of open-weight models?
Open models can be misused for malicious purposes, such as generating disinformation, surveillance, or automated deception. Responsible use policies are necessary to mitigate these risks, but enforcement remains a challenge.
What does this mean for future AI development?
This release signals a shift toward greater transparency and ownership in AI, potentially encouraging more open models. However, it also raises questions about balancing openness with safety and control in sensitive applications.
Source: ThorstenMeyerAI.com
