📊 Full opportunity report: The Significance Of Thinking Machines’ Hints In AI Development on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released its large-scale AI model, Inkling, openly on Hugging Face under Apache 2.0, signaling transparency and a shift in open AI development. The release highlights industry practices and raises questions about open source claims.
Thinking Machines has publicly released its flagship AI model, Inkling, on Hugging Face under the Apache 2.0 license. This marks a notable shift in how large language models are shared, emphasizing transparency and ownership, and directly addresses industry questions about open-source claims and model control.
The Inkling model is a 975-billion-parameter, mixture-of-experts transformer supporting multimodal input (text, images, audio) with a 1-million-token context window. It was trained on 45 trillion tokens across various modalities, using a hybrid optimizer and over 30 million reinforcement learning rollouts. The full weights are now available on Hugging Face, accompanied by detailed specifications and performance metrics.
In addition to the large model, a smaller variant, Inkling-Small, with 276 billion parameters and 12 billion active experts, was also previewed, showing competitive benchmark results. The release is accompanied by a candid discussion of training methods, including synthetic data use, and emphasizes the importance of owning and controlling models rather than renting access through APIs.
However, the release also included a notable caveat: while the weights are open under Apache 2.0, the training data and full pipeline are not published, and there are reports of a separate Model Acceptable Use Policy (AUP) that restricts certain uses, such as surveillance and automated decision-making, which could complicate open-source claims.
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.
Industry Shift Toward Transparent, Open-Weight Models
This release signals a potential shift in AI development, emphasizing transparency, ownership, and control over large models. By providing open weights and openly discussing training practices, Thinking Machines challenges the industry norm of proprietary models and API-based access, possibly influencing future open-source AI releases.
It also raises important questions about the true nature of open source in AI, especially concerning licensing restrictions and use policies layered on top of open weights. For organizations in sensitive domains, this could impact deployment decisions and regulatory considerations.
LAFVIN AI Chatbot Kit for ESP32-S3, Preloaded OpenAI & Deepseek Voice Assistant Projects, Voice Wake-up & Real-time Interruption, Suitable for Learning AI and IoT Projects.
- Powerful ESP32-S3 Processor: Dual-core Xtensa 32-bit LX7, 512KB SRAM
- Preloaded AI Voice Projects: Deepseek and OpenAI dialogue platforms included
- Stable Wireless Connectivity: Wi-Fi 2.4GHz and Bluetooth 5 (LE)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Recent Trends in Open-Source AI and Industry Practices
Over the past year, there has been increasing debate over the openness of large language models, with some companies releasing models behind closed APIs and others sharing weights openly. Earlier efforts, like Meta’s Llama and EleutherAI’s models, set precedents for open weights, but many proprietary models remain closed or restricted.
Thinking Machines’ approach, combining open weights with transparency about training data and policies, represents a nuanced position that balances openness with responsible use. The recent release comes amid broader industry discussions on AI safety, ownership, and regulatory frameworks.
Notably, the model’s release follows a period of heightened scrutiny after government directives to switch off certain models, emphasizing the importance of owning and controlling AI assets rather than relying solely on external APIs.
“We believe owning your model is crucial for responsible AI development. Our release aims to foster transparency and innovation.”
— Thinking Machines spokesperson
LLM Systems Engineering: Training and Building Large Language Models – Engineering AI Models Through Fine-Tuning, Continued Pretraining, and From-Scratch Development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Open Source Claims vs. Use Restrictions
It remains unclear how the separate Model Acceptable Use Policy (AUP) will be enforced and whether it will significantly restrict the open-source nature of the weights. The exact scope of restrictions and their legal enforceability are still under question, requiring further clarification from Thinking Machines.
HIWONDER Humanoid Robot with ChatGPT Multimodal AI Models AI Embodied Intelligent Vision Scene Voice Understanding 18DOF Educational Robot Kit Python Programming, TonyPi Advanced & RaspberryPi 5 8GB
- Powered by Raspberry Pi 5: High-performance AI vision robot
- Open-source development platform: Supports advanced AI robotics development
- ChatGPT multimodal integration: Enhanced human-machine interaction
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps in Industry Adoption and Policy Clarification
Expect further analysis and independent benchmarking of Inkling’s performance and licensing terms. Industry observers will scrutinize the AUP and its implications for open-source AI development. Additionally, other organizations may follow suit, releasing models with similar transparency and restrictions.
Regulatory bodies and user communities are likely to monitor how these layered policies impact responsible AI deployment and ownership.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What makes Inkling different from other large language models?
Inkling is notable for its open weights under Apache 2.0, its multimodal capabilities, and its emphasis on model ownership, contrasting with many proprietary models that are only accessible via APIs.
What are the potential risks of layered use policies on open models?
Such policies could limit responsible use, create legal uncertainties, and undermine the transparency that open weights aim to provide, especially if enforcement is inconsistent.
How might this release influence future AI model sharing?
It could encourage more organizations to share weights openly while implementing responsible use policies, balancing transparency with safety considerations.
Is the training data for Inkling publicly available?
No, the training data and full training pipeline have not been published, which is typical in the industry but raises questions about reproducibility and transparency.
Source: ThorstenMeyerAI.com
