📊 Full opportunity report: What Early Hints From Thinking Machines Tell Us About AI’s Growth on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large open-weight AI model, and shared early performance data. This offers new insights into AI’s growth and open-source trends, but some details remain unclear.
Thinking Machines has released its first foundation model, Inkling, with full weights available openly on Hugging Face. This marks a notable shift in AI deployment practices, emphasizing transparency and ownership over rented models, and provides early benchmarks on its performance.
Inkling is a 975-billion-parameter multimodal transformer trained on 45 trillion tokens, supporting a one-million-token context window. Its weights are released under Apache 2.0 license, allowing download, modification, and deployment on private infrastructure. Unlike typical industry practices, the company explicitly states that Inkling is not the strongest model available, acknowledging its limitations.
Details about the training process include the use of hybrid optimizers and over 30 million reinforcement learning rollouts, with some testing conducted on synthetic data generated by open-weight models like Kimi K2.5. Early benchmark results show strong performance in speech and safety metrics but middling scores in some language understanding tasks.
While the open weights are a significant step, questions remain about the full scope of the licensing restrictions and the company’s separate Model Acceptable Use Policy, which reportedly forbids surveillance and deception, potentially complicating the open-source narrative.
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 Development
This release demonstrates a shift toward more transparent AI development, allowing organizations to own and modify models directly rather than relying solely on API access. It signals a move toward democratizing AI technology, enabling wider experimentation and deployment.
However, the presence of a separate use policy raises questions about the true openness of Inkling, especially for applications in sensitive domains. The early performance data provides a benchmark for the industry but also highlights the ongoing trade-offs between model size, safety, and capabilities.
Accelerate Everything with Tensor Cores: A Developer’s Guide to High-Performance AI, Efficient Training, and Scalable Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Recent Trends in Open-Source AI Models
Over the past year, several AI labs have begun releasing larger models with open weights, challenging the dominance of closed, API-only models. Notably, Meta and EleutherAI have contributed to this trend, emphasizing transparency and community collaboration. Thinking Machines’ release of Inkling follows this pattern but is distinguished by its candid acknowledgment of its limitations and its licensing approach.
This development occurs amid ongoing debates about responsible AI use, licensing restrictions, and the balance between openness and safety. The release of Inkling’s full weights under an open license marks a significant milestone in these discussions.
“Releasing the full weights under Apache 2.0 is a game-changer, but the accompanying use restrictions complicate the narrative of true openness.”
— Thorsten Meyer, AI researcher
high performance external hard drives for AI data
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Inkling’s Use Policies
It remains unclear how the separate Model Acceptable Use Policy will be enforced and whether it will limit the practical openness of Inkling in sensitive applications. The exact scope of restrictions and their enforceability are still unverified, raising questions for potential users.
Additionally, the full details of the training data and pipeline have not been disclosed, leaving some uncertainty about the model’s transparency and reproducibility.
multimodal transformer AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps in Evaluating Inkling and Open Models
Further independent benchmarking and testing will be necessary to verify Inkling’s claimed performance, especially in real-world scenarios. Companies and researchers will likely examine the licensing restrictions closely before adopting the model for sensitive or regulated domains.
Expect more open models to follow, with increased emphasis on transparency about licensing and use policies, shaping the future landscape of accessible AI technology.
AI development books for programmers
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 notable for its full open weights under Apache 2.0 license, allowing users to download, modify, and deploy it privately. It also emphasizes ownership and transparency, though it reportedly includes a separate use policy that may impose restrictions.
Does open-weight mean the model is fully open source?
Not necessarily. While the weights are open under Apache 2.0, the training data, pipeline, and use restrictions are not fully disclosed, which complicates the open-source classification.
What are the main performance strengths of Inkling?
Early benchmarks show strong results in speech recognition, safety, and certain reasoning tasks, but it is not the top-performing model in all language understanding benchmarks.
What are the potential risks of using Inkling?
Potential risks include restrictions imposed by the separate use policy, which could limit certain applications, especially in sensitive or regulated sectors. Enforcement and scope of these restrictions remain unclear.
Source: ThorstenMeyerAI.com
