📊 Full opportunity report: Interpreting The Inkling Of Thinking Machines For AI Innovation on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released its first foundation model, Inkling, with open weights under Apache 2.0. The company openly states it is not the strongest model available, highlighting transparency in its launch.
Thinking Machines has officially released the open weights of its new foundation model, Inkling, under the Apache 2.0 license, making it accessible for download, modification, and deployment. This marks a notable move in the AI community, emphasizing transparency and ownership over proprietary models, and comes amidst ongoing debates about open-source AI development and licensing restrictions.
Inkling is a 975-billion-parameter Mixture-of-Experts transformer supporting multimodal inputs—text, images, and audio—processed jointly without a dedicated vision adapter. It was pretrained on 45 trillion tokens of diverse data, including text, images, audio, and video, and features a 1-million-token context window. The model is designed for open use, with weights available on Hugging Face under Apache 2.0, allowing users to fine-tune, modify, and deploy independently.
Thinking Machines explicitly stated that Inkling “is not the strongest model available today,” and the launch included candid details about training methods, including the use of synthetic data generated by open-weight models like Kimi K2.5 during post-training bootstrap. The company also indicated it maintains a separate Model Acceptable Use Policy (AUP), which restricts surveillance, deception, and fully automated decision-making affecting individuals’ rights—raising questions about the true openness of the model.
While the open weights are freely available, the full training data and pipeline are not published, a common industry practice. The company’s transparency about performance benchmarks shows strengths in areas like speech and safety but acknowledges middling results in text-only tasks. This release raises important questions about licensing, restrictions, and the model’s actual openness.
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 Model Release
The release of Inkling under an open license marks a significant shift toward transparency and user ownership in AI development. It provides developers and organizations with direct access to a large, multimodal foundation model, enabling independent fine-tuning and deployment without reliance on proprietary APIs. This could accelerate innovation, reduce vendor lock-in, and promote open research. However, the presence of a separate AUP and the lack of full training data introduce complexities around the true openness and permissible uses of the model, especially in sensitive domains like surveillance and decision-making.
This move also signals a response to recent industry and regulatory debates about the ethics and control of powerful AI models, emphasizing that openness can coexist with restrictions. The model’s performance benchmarks suggest it is competitive but not the top-tier, reinforcing the idea that transparency and honesty about capabilities are becoming more valued than just raw power.
open source AI model download
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Open AI Model Releases
Over recent years, the AI community has seen a tension between proprietary models and open-source initiatives. Leading companies like OpenAI initially released models with restricted access, citing safety and misuse concerns, but have gradually moved toward more open practices. The release of models like Meta’s Llama and Meta’s Llama 2 under open licenses has sparked a broader shift toward transparency and user control.
Thinking Machines, founded by former OpenAI CTO, has built a reputation for transparency and ambitious open releases. Its recent launch of Inkling, with open weights and detailed performance benchmarks, continues this trend, positioning itself as a challenger to proprietary models while openly acknowledging its limitations in performance compared to the state-of-the-art.
Industry observers note that the combination of open weights with a restrictive AUP complicates the narrative of true open source, raising questions about enforcement and scope, especially in sensitive applications like surveillance or automated decision-making.
ESP32 Basic Starter Ai Chatbot Kit Development Board USB-C Dual Core Microcontroller Support AP/STA/AP+STA Compatible with Arduino IDE IoT for Beginners Engineers
【High Performance】The ESP32 module is equipped with a dual-core CPU and features a Type-C USB interface, with 44…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Inkling’s Openness
It remains unclear how strictly the Model Acceptable Use Policy will be enforced and whether it effectively limits misuse in practice. The full training data and pipeline are not published, raising questions about reproducibility and transparency. Additionally, the extent to which the model can be freely modified or commercialized, given potential restrictions, is still uncertain.
Further clarity is needed on how the licensing and AUP interact, especially in sensitive domains such as surveillance or automated decision-making, where restrictions could impact deployment and compliance.
AI model fine-tuning software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Evaluation
Expect independent researchers and organizations to evaluate Inkling’s performance across various benchmarks and real-world tasks. Attention will focus on how the AUP influences practical use, especially in regulated or sensitive sectors. Thinking Machines is likely to release more detailed documentation and possibly clarify enforcement of restrictions.
Further testing and comparative analysis will determine if Inkling can serve as a viable alternative to proprietary models, and whether its open approach influences industry standards for transparency and responsible AI deployment.
large language model training tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Is Inkling truly open source?
The weights are released under Apache 2.0, allowing modification and deployment, but the full training data, pipeline, and potentially restrictive use policies are not fully disclosed, complicating the definition of ‘truly open.’
What are the main restrictions on Inkling’s use?
According to reports, a separate Model Acceptable Use Policy prohibits surveillance, deception, and fully automated decisions affecting individuals’ rights, which may limit how the model can be used in practice.
How does Inkling compare to other models in performance?
Benchmarks show strengths in speech and safety but middling results in text-only tasks. It is not the top-performing model but offers a transparent, open-weight alternative for many applications.
Will the full training data be released?
No, the training data and full pipeline have not been published, which limits full reproducibility and transparency, common in industry but a point of concern for some researchers.
What does this mean for the future of open AI models?
This release signals a move toward more transparent, owner-controlled models, but also highlights ongoing debates about restrictions, licensing, and responsible use in AI development.
Source: ThorstenMeyerAI.com
