📊 Full opportunity report: What Thinking Machines’ Inkling Signals About AI’s Evolution on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has publicly released its Inkling model as open weights under Apache 2.0, openly stating it is not the top-performing model. This move emphasizes transparency in AI development and ownership. The implications for AI openness and licensing are significant but raise questions about restrictions and data transparency.
Thinking Machines has publicly released its first foundation model, Inkling, as open weights on Hugging Face under the Apache 2.0 license, making it freely downloadable and modifiable. The company explicitly stated that Inkling is not the strongest model available today, signaling a shift towards transparency and ownership in AI development.
Inkling is a Mixture-of-Experts transformer with 975 billion total parameters and 41 billion active, supporting a 1-million-token context window. It was pretrained on 45 trillion tokens across various modalities, including text, images, audio, and video. The model is multimodal on input, processing text, images, and audio jointly without an encoder, with components trained from scratch.
The full weights were released first on Hugging Face, with day-zero support in several open-source frameworks like transformers, vLLM, SGLang, and llama.cpp. This contrasts with typical industry practice where models are often released as closed or with limited access. The release was accompanied by transparency about the model’s capabilities, training data, and licensing, emphasizing user ownership and control.
However, there are important caveats: the weights are under Apache 2.0, but the training data and pipeline remain undisclosed. Additionally, reports suggest that Thinking Machines maintains a separate Model Acceptable Use Policy (AUP) that restricts certain uses like surveillance and deception, which could complicate the open-source framing. The company also previewed a smaller version, Inkling-Small, with promising benchmark results, which will be released after further testing.
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
This release marks a notable shift in AI development, emphasizing model ownership and transparency. By providing open weights under a permissive license and openly stating that Inkling is not the top model, Thinking Machines challenges the industry norm of proprietary models and signals a move towards more accessible AI tools.
For developers and organizations, this means greater control over models, including the ability to fine-tune, inspect, and deploy independently. It also raises questions about licensing restrictions, especially considering the reported separate AUP that limits certain use cases, which could influence how the model is adopted in sensitive domains.
Overall, this development could accelerate innovation and democratize access but also prompts careful consideration of licensing and ethical restrictions.
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.
Background on AI Model Releases and Industry Norms
In recent years, most large foundation models have been released as closed or with limited access, often with proprietary licenses that restrict modification and deployment. Open-source releases like Meta’s Llama or EleutherAI’s models have aimed to democratize AI, but many models remain closed or partially accessible.
Thinking Machines, founded by former OpenAI CTO Thorsten Meyer, has a reputation for transparency and innovation. Its decision to release Inkling openly, while openly acknowledging it is not the strongest model, aligns with broader industry discussions about model ownership, licensing, and the ethics of AI deployment. The move also responds to recent debates about the risks of opaque models and the importance of user control.
Historically, the industry has seen a tension between commercial interests and open development. Inkling’s release under Apache 2.0, combined with the explicit statement about its capabilities, marks a significant moment in this ongoing debate.
“Releasing Inkling openly, even while acknowledging it’s not the top model, is a step towards greater transparency and ownership in AI.”
— Thorsten Meyer, founder of Thinking Machines
multimodal AI development tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Licensing and Data Transparency
It remains unclear how Thinking Machines’ separate Model Acceptable Use Policy (AUP) will be enforced and how it interacts with the Apache 2.0 license. The training data and pipeline are not disclosed, raising questions about data transparency and reproducibility. The extent to which users can freely modify and commercialize the model without restrictions is still uncertain, especially given the reported restrictions in the AUP.
open source AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Evaluation of Inkling
Further testing and independent benchmarking of Inkling and Inkling-Small are expected, with full weights and detailed documentation to follow. Industry observers will monitor how organizations adopt the model, especially in sensitive domains, and how Thinking Machines enforces its AUP. The broader impact on open-source AI development will become clearer as more users experiment with the model.
AI model licensing and licensing management software
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 model released openly under Apache 2.0, with a focus on transparency and ownership, unlike many proprietary models.
Can I modify and commercialize Inkling freely?
Under the Apache 2.0 license, you can modify and commercialize the weights, but the reported separate AUP may impose restrictions on certain uses, which should be reviewed before deployment.
Why is the model not considered the strongest available?
Thinking Machines explicitly stated that Inkling is not the strongest model currently, prioritizing transparency and open access over raw performance.
What are the ethical considerations of this open release?
The model’s licensing and use restrictions, along with undisclosed training data, raise questions about ethical use, data bias, and enforceability of restrictions.
Source: ThorstenMeyerAI.com
