📊 Full opportunity report: The Weights Came First: What Thinking Machines’ Inkling Actually Signals on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has publicly released the full weights of its foundation model Inkling under an open license, marking a notable shift in AI model deployment. This move emphasizes transparency and ownership over renting access, but raises questions about restrictions and data sources.
Thinking Machines has publicly released the full weights of its foundation model Inkling on Hugging Face under the Apache 2.0 license, making it the first major lab to do so before releasing performance benchmarks or closed API access. This move signals a shift toward transparency and ownership in the AI model landscape, with potential implications for how companies deploy and control large language models.
Inkling is a 975-billion-parameter, mixture-of-experts transformer supporting a 1-million-token context window, trained on 45 trillion tokens across multiple modalities including text, images, audio, and video. It features a native multimodal input design, processing audio as spectrograms and images as pixel patches, trained from scratch without vision adapters.
The model’s weights were released under the Apache 2.0 license, allowing download, modification, and commercial use. This contrasts with typical industry practice, where models are often released with restrictions or only via API access. The release also included a preview of Inkling-Small, a 276-billion-parameter variant, which reportedly matches or exceeds certain benchmarks.
However, the lab clarified that the weights are not open source — the training data and full training pipeline are not published. Additionally, reports indicate that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP) that restricts surveillance, deception, and automated decision-making, potentially complicating the open licensing framework. The model’s performance claims include high scores on certain benchmarks like AIME 2026 (97.1%) and VoiceBench (91.4%), but mixed results on others.
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 Before Benchmarking
This release marks a significant departure from the industry norm of releasing models via API or with limited access, emphasizing ownership and transparency. It allows organizations to fine-tune, inspect, and deploy the model independently, potentially accelerating AI innovation and democratization.
However, the presence of a separate AUP and the lack of full training data raise questions about restrictions on use and reproducibility. The move could influence future model releases, pushing other labs toward more open practices, but also highlighting ongoing tensions between openness and control in AI development.
HP 15.6" Laptop – Complete Productivity Solution, Windows 11, Microsoft Office, Intel 4 Core N100, 16GB RAM, 640GB Storage (128GB UFS + 512GB SD Card), Copilot AI, Lightweight – Silver
Upgraded to 16GB High-Speed Memory Boosted to 16GB RAM for improved multitasking — switch between apps, browser tabs,…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background on Inkling and Industry Practices
Thinking Machines, founded 17 months ago by former OpenAI CTO Greg Brockman and staffed with engineers involved in ChatGPT’s development, has been known for its focus on safety and transparency. Prior to this release, most foundation models from leading labs were either accessible only via APIs or released with limited open weights, often accompanied by restrictions.
The recent trend in AI has been toward closed, proprietary models, partly driven by commercial interests and safety concerns. Inkling’s open release under Apache 2.0, combined with its candid performance disclosures, signals a potential shift in how foundational models are shared and owned. The release occurred shortly after a major incident where a frontier model was switched off globally by authorities, intensifying debates over model control and data sovereignty.
Previously, the industry has grappled with balancing transparency, safety, and commercial viability, with most labs opting for controlled releases. Inkling’s approach challenges this paradigm, emphasizing user ownership but also raising questions about restrictions and data provenance.
“We believe in empowering users with full control over our models, which is why we released Inkling’s weights under Apache 2.0.”
— Thinking Machines spokesperson
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 Licensing and Use Restrictions
It remains unclear how the separate Model Acceptable Use Policy (AUP) interacts with the Apache 2.0 license, and whether restrictions could limit full ownership or commercial deployment. The exact scope and enforceability of this policy are not publicly verified, raising questions for organizations considering adopting Inkling.
Additionally, details about the training data, including data sources and preprocessing steps, have not been disclosed, which affects assessments of model bias and reproducibility. The performance benchmarks are preliminary, with full results and comparisons still pending.
It is also uncertain whether other labs will follow suit in releasing models openly before benchmarking or if this is a unique case driven by specific strategic considerations.
NVIDIA Certified Associate: Generative AI LLMs (NCA-GENL) (NVIDIA Certification Guides)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Benchmarking
Following this release, independent researchers and organizations will likely test Inkling’s performance across diverse tasks, verify claims, and evaluate the legal restrictions imposed by the AUP. The full benchmark results for Inkling-Small are expected soon, providing more clarity on its capabilities.
Thinking Machines may also clarify the scope of its licensing and restrictions, potentially issuing updated policies or documentation. Other AI labs might reconsider their release strategies, balancing openness with safety and control concerns.
In the coming months, the AI community will observe whether this open-weight approach influences broader industry practices and how it impacts the development of truly open, ownership-based AI models.
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 does the open release of Inkling’s weights mean for AI developers?
It allows developers to download, modify, and deploy the model independently, fostering innovation and customization without relying on API access.
Are there restrictions on how Inkling can be used?
Yes, according to reports, Thinking Machines maintains a separate Acceptable Use Policy that restricts surveillance, deception, and certain automated decision-making, which may limit some applications despite the open weights.
Why did Thinking Machines release the weights before benchmarking?
The company aims to promote transparency and ownership in AI, believing that open weights empower users and accelerate responsible AI development.
What are the risks of releasing open weights without full training data?
Risks include reproducibility issues, potential bias, and lack of transparency about data sources, which can affect trust and safety assessments.
What happens next after this release?
Expect independent testing of Inkling’s performance, clarification of license restrictions, and potential shifts in industry practices toward more open model sharing.
Source: ThorstenMeyerAI.com
