📊 Full opportunity report: AI Breakthroughs: Understanding What Thinking Machines’ Inkling Means on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has publicly released the Inkling model, a large, open-weight multimodal transformer. This move emphasizes transparency and ownership over proprietary models, but some restrictions remain. The development signals a shift toward more open AI ecosystems.
Thinking Machines has publicly released its latest multimodal foundation model, Inkling, under the Apache 2.0 license. This marks a notable shift in the AI landscape, emphasizing model ownership and transparency over proprietary control. The release includes full model weights available on Hugging Face, making Inkling accessible for download, modification, and deployment.
Inkling is a 975-billion-parameter mixture-of-experts transformer supporting a 1-million-token context window. It was trained on 45 trillion tokens across text, images, audio, and video, with a native multimodal input design that processes audio as spectrograms and images as pixel patches, all trained from scratch without vision adapters. The model’s weights are released openly under Apache 2.0, allowing broad use and customization.
The company, Thinking Machines, explicitly stated that Inkling “is not the strongest model available today, closed or open”, prioritizing openness and transparency. They also disclosed training details, including the use of external benchmarks and reinforcement learning, with some testing conducted on synthetic data generated by other open-weight models like Kimi K2.5. However, questions remain regarding the full scope of licensing restrictions, as reports suggest a separate Acceptable Use Policy may impose limitations on surveillance, deception, and automated decision-making.
While the model’s open weights are a significant step, critics highlight that the training data and pipeline remain proprietary, and the licensing restrictions may influence how the model can be used, especially 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 Release for AI Ecosystems
The release of Inkling under an open license challenges traditional proprietary AI models by providing full access to weights and training data, fostering innovation, and enabling broader ownership. This move aligns with a growing push for transparency and control in AI development, potentially accelerating adoption in sectors like research, industry, and public safety.
However, the presence of a separate Acceptable Use Policy raises questions about the scope of openness and whether restrictions could limit certain applications, especially in areas like surveillance or automated decision-making. The move also signals a strategic shift, emphasizing ownership over rental models and possibly influencing future industry standards.
AI Agents for Developers with Python and MCP: Build Production-Ready Agents with LangGraph, RAG, Tool Use, Multi-Agent Workflows, FastAPI, and AgentOps (Production AI Engineering Series Book 1)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Open-Weight Model Releases
Historically, most large AI models have been released with closed weights or limited access, prioritizing commercial control and safety. Recent efforts, including Meta’s Llama 2 and OpenAI’s open initiatives, have begun to shift toward more transparent models. Thinking Machines’ decision to release Inkling openly, with full weights and detailed training information, represents a notable development in this trend.
Previous releases often involved restrictions through licenses or API controls, but Inkling’s open weights under Apache 2.0 mark a significant departure, emphasizing user ownership and customization. The model’s design as a multimodal transformer trained on diverse data sets further underscores the trend toward versatile, accessible AI systems.
Despite this progress, concerns about proprietary training data, licensing restrictions, and safety policies continue to shape the conversation around open AI models.
“We believe in empowering the community with full access to our models, and Inkling is a step toward more open, accountable AI development.”
— Thinking Machines spokesperson
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Licensing Restrictions and Usage Limitations
While the weights are openly available under Apache 2.0, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy that may impose restrictions on surveillance, deception, and automated decision-making. The exact scope and enforceability of this policy are not publicly verified, raising questions about the true openness of the model’s use.
It remains unclear how these restrictions will impact commercial or sensitive applications, and whether they could limit the model’s deployment in certain sectors.
From Weights to Wisdom: The Complete Guide to Running and Adapting Opensource AI Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Monitoring Adoption and Policy Clarifications
Expect further clarification from Thinking Machines regarding the scope of their Acceptable Use Policy and how it interacts with the open weights license. Industry observers will also watch for independent benchmarking and real-world applications to assess the model’s performance and safety.
Additionally, other organizations may follow suit, releasing their own models openly or with new licensing frameworks, shaping future AI development standards.
Fine-Tuning Large Language Models: From Custom Datasets to High-Performance AI Models Using Modern Toolchains
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 multimodal transformer with native support for text, images, and audio, and it is released under an open Apache 2.0 license, allowing broad access and modification.
Are there any restrictions on how I can use Inkling?
While the weights are openly available, reports suggest a separate Acceptable Use Policy may impose restrictions on surveillance, deception, and automated decision-making. Users should verify the policy before deploying the model in sensitive applications.
Why is the open licensing of Inkling significant?
Open licensing allows users to download, modify, and deploy the model independently, promoting transparency, innovation, and control—challenging the traditional proprietary AI ecosystem.
What are the potential risks of open weights without full transparency?
Risks include the use of the model in harmful applications, unverified safety measures, and ambiguity around licensing restrictions. Full transparency on training data and policies is essential for responsible deployment.
What is the next step for the AI community regarding Inkling?
Monitoring how organizations adopt and adapt Inkling, clarifying licensing restrictions, and benchmarking its performance in real-world scenarios will shape its impact and future developments.
Source: ThorstenMeyerAI.com
