📊 Full opportunity report: The Impact Of Thinking Machines’ Inkling On AI’s Future Path on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large, open-weight AI model under Apache 2.0 license, openly stating it is not the top performer. This move emphasizes transparency and ownership in AI development, but raises questions about licensing and use restrictions.
Thinking Machines has released its first foundation model, Inkling, under an open-source license, making it publicly available on Hugging Face. The company explicitly stated that Inkling is not the strongest model on the market, marking a departure from typical industry practice of promoting top performance over transparency. This move highlights a shift toward prioritizing open access and ownership in AI development, which could influence future industry standards and practices.
Inkling is a Mixture-of-Experts transformer with 975 billion parameters, capable of processing multimodal inputs—text, images, and audio—without relying on vision adapters. It was trained on 45 trillion tokens and supports a 1-million-token context window. The model’s weights are released under Apache 2.0 license, enabling download, modification, and commercial use, which is a notable shift toward open ownership. The training involved hybrid optimization methods and synthetic data from open-weight models, including Chinese model Kimi K2.5.
In addition to Inkling, a smaller variant, Inkling-Small, with 276 billion parameters, was previewed and shown to match or surpass larger models on several benchmarks, though full weights for this version are pending. The company’s transparency about performance and licensing is unusual in the context of large language models, which are often proprietary.
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, coupled with the admission that it is not the strongest model, signals a possible shift in industry norms. It emphasizes ownership, transparency, and control over AI models, especially after recent incidents of model shutdowns due to government directives. However, the existence of a separate Acceptable Use Policy raises questions about restrictions and enforceability, which could influence how organizations adopt and trust open models.
This move could encourage more companies to release models openly, fostering a more collaborative and transparent AI ecosystem, but also prompts scrutiny over licensing terms and usage restrictions that may limit practical deployment.
Large Language Models: The Hard Parts: Open Source AI Solutions for Common Pitfalls
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and Recent Open-Model Trends
Historically, most large foundation models have been released as proprietary, with limited access to weights and training data. When open, they often lack transparency regarding licensing and usage policies. Recent incidents, such as government-ordered shutdowns, have heightened interest in models that can be owned and operated independently.
Thinking Machines’ decision to publish Inkling’s weights openly, alongside a candid performance report, marks a notable departure from this norm. The company’s emphasis on transparency and ownership aligns with broader industry discussions about control, safety, and responsible AI deployment.
“We believe in giving the community access to powerful models with clear licensing, even if they are not the absolute best. Ownership matters.”
— Thinking Machines spokesperson
Build Your Own Language Model: From Raw Text and Tokenizers to a Safe, Tool-Using Multimodal AI Assistant (Made Simple AI Series Book 3)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Licensing and Use Restrictions
It is not yet clear how the separate Model Acceptable Use Policy interacts with the Apache 2.0 license, especially regarding restrictions on surveillance, deception, and automated decision-making. The enforceability and scope of these restrictions remain unverified, which could impact how organizations adopt Inkling for sensitive applications.
Further clarification is needed on whether the AUP is legally binding or merely a guideline, and how it might influence commercial or research use.
How LLMs Actually Work: Large Language Models Explained: From Hardware to Hallucination, Tokens to Transformers
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 detailed analysis and independent testing of Inkling’s performance and licensing restrictions. Industry stakeholders will scrutinize the AUP, and more organizations may follow with open releases emphasizing ownership. Regulatory bodies and user communities will likely monitor how licensing and restrictions evolve, influencing future open-source AI practices.
Further releases, including full weights for Inkling-Small, are anticipated, alongside ongoing benchmarking and safety assessments to validate the model’s capabilities and compliance.
Foundations of Building Custom AI Models: A Practical Guide to Understanding AI, LLM Architecture, and Dataset Design (Mastering Custom AI Systems Book 1)
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 openly available under the Apache 2.0 license, supports multimodal inputs, and explicitly states it is not the strongest model, emphasizing transparency and ownership over performance supremacy.
Does open weights mean the model is fully open source?
No. The weights are under Apache 2.0 license, but the training data, pipeline, and possibly usage restrictions via an Acceptable Use Policy are not fully disclosed, which limits true open source status.
What are the potential risks of using Inkling?
Risks include uncertainties about licensing restrictions, enforceability of the Acceptable Use Policy, and whether the model can be safely used in sensitive applications without hidden limitations.
Why does this release matter for the AI industry?
It signals a shift toward prioritizing model ownership and transparency, which could influence future industry norms and foster more open collaboration, but also raises questions about licensing and restrictions.
Source: ThorstenMeyerAI.com
