📊 Full opportunity report: Cracking The Code: What Thinking Machines’ Inkling Indicates For AI on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large open-weight AI model with 975 billion parameters, openly acknowledging it is not the strongest available. The release emphasizes transparency and open access, raising questions about licensing and use policies.
Thinking Machines has released its first foundation model, Inkling, a 975-billion-parameter, open-weight, multimodal transformer, openly available on Hugging Face. The company emphasized that Inkling is not the strongest model available, marking a significant departure from typical industry practices of claiming market dominance.
Inkling is a Mixture-of-Experts transformer supporting a 1-million-token context window and trained on 45 trillion tokens across text, images, audio, and video. Its architecture routes tokens through 66 layers with 41 billion active parameters, and it supports multimodal input—text, images, and audio—processed jointly without additional vision adapters. The model’s weights are released under Apache 2.0 license on Hugging Face, enabling free download, modification, and commercial use.
In addition, a smaller variant, Inkling-Small, with 276 billion total parameters and 12 billion active, has been previewed. Its performance, thanks to improved training techniques, matches or exceeds larger models on various benchmarks. The training process involved hybrid optimizers, reinforcement learning, and synthetic data generated by other open models. The release includes detailed performance scores, with notable strengths in speech and safety benchmarks, but moderate results in some language understanding tasks.
While the weights are openly available, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP), which may impose restrictions on surveillance, deception, and automated decision-making, potentially complicating the ‘open’ nature of the release. The company clarified that the weights are not open source in the traditional sense, as the training data and pipeline are not publicly disclosed.
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.
Impact of Transparency and Open Access in AI
This release underscores a shift toward transparency in the AI industry, with Thinking Machines openly sharing weights and acknowledging its model’s limitations. Such honesty can influence industry standards, encouraging other developers to be more transparent about their models’ strengths and weaknesses. It also raises important questions about licensing, restrictions, and the balance between open access and responsible use, especially in sensitive domains like surveillance and automated decision-making.
For users and developers, the open weights mean greater control and flexibility, enabling local deployment and customization without reliance on proprietary APIs. However, the potential restrictions outlined in the company’s AUP could limit certain applications, making it crucial for users to understand the legal and ethical boundaries before deploying Inkling.
AI model training software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Trends Toward Open-Weight Models and Transparency
Over the past year, the AI industry has seen a growing emphasis on open-sourcing models, driven by calls for transparency and democratization. Major players like Meta and EleutherAI have released large models under permissive licenses, emphasizing community-driven development. However, most releases are accompanied by caveats regarding training data, licensing restrictions, or safety policies. Thinking Machines’ approach—releasing weights openly but clarifying limitations—fits within this broader movement towards transparency, contrasting with closed, API-only models from companies like OpenAI and Anthropic. The timing coincides with increased regulatory scrutiny and societal debates over AI safety, bias, and misuse.
Previously, the industry has grappled with issues of model transparency, reproducibility, and responsible deployment. Inkling’s release, with candid performance metrics and open weights, signals a possible trend toward more honest disclosures, even if restrictions remain layered on top.
“We believe in open access and responsible use. Our AUP reflects our commitment to ethical AI deployment.”
— Thinking Machines spokesperson
multimodal AI development kits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unanswered Questions About Licensing and Use Restrictions
While the weights are openly available under Apache 2.0, reports suggest that Thinking Machines enforces a separate Model Acceptable Use Policy (AUP) that may restrict surveillance, deception, and certain automated decision-making applications. The exact scope, enforceability, and legal implications of this policy are not yet publicly confirmed, raising questions about how ‘open’ the model truly is for different use cases.
It remains unclear whether the AUP will be strictly enforced or if it primarily serves as a guideline, and how it compares to traditional open-source licenses. Additionally, the impact of these restrictions on commercial and research use is still to be determined.
large open-weight AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Community Engagement
Following the release, attention will turn to third-party testing and independent benchmarking of Inkling’s performance and safety. Developers and organizations will likely experiment with fine-tuning and deploying the model in various domains, especially in speech and multimodal applications. Monitoring how Thinking Machines enforces its AUP and whether the community adopts the model responsibly will be crucial.
Further updates are expected as the company releases the full weights of Inkling-Small and provides more detailed performance data. Industry observers will also watch for how competitors respond, whether through open releases or more guarded approaches.
AI model licensing and deployment tools
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 a 975-billion-parameter, open-weight, multimodal transformer openly available under Apache 2.0, and it openly admits that it is not the most powerful model currently on the market.
Are the weights truly open source?
The weights are released under Apache 2.0, allowing free use and modification, but the training data and pipeline are not publicly disclosed. Additionally, a separate Acceptable Use Policy may impose restrictions on deployment.
What are the main strengths of Inkling?
It shows strong performance in speech and safety benchmarks, supports multimodal input, and offers a large context window, making it versatile for various applications.
Will the restrictions limit its use?
Potential restrictions in the Model Acceptable Use Policy could limit applications involving surveillance, deception, or automated decision-making, but the full scope is not yet clear.
Source: ThorstenMeyerAI.com
