📊 Full opportunity report: Is Thinking Machines’ Inkling A Sign Of Major AI Progress? on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has released Inkling, a large open-weight multimodal AI model. While it is not the strongest available, its open release and transparency mark a notable development in AI progress and open-source practices.
Thinking Machines has publicly released the full weights of its new AI model, Inkling, under an open-source license, marking a significant moment in AI development. This move comes amid ongoing debates about transparency, ownership, and the pace of AI progress, and is notable because the model is openly available but not claimed to be the most powerful today.
The Inkling model is a 975-billion-parameter mixture-of-experts transformer supporting multimodal inputs, including text, images, and audio. It was pretrained on 45 trillion tokens of diverse data and supports a 1-million-token context window. Unlike many recent models, its full weights were released immediately on Hugging Face under the Apache 2.0 license, allowing users to download, modify, and deploy independently.
According to the announcement, Inkling is not the strongest model available, but its open release and transparent methodology represent a noteworthy shift. The model’s training involved hybrid optimization techniques and over 30 million reinforcement-learning rollouts, with testing data indicating competitive performance in speech and safety benchmarks. However, its scores on some benchmarks, like Humanity’s Last Exam, remain mid-range compared to larger closed models.
Additionally, the company reportedly maintains a separate Model Acceptable Use Policy that restricts surveillance, deception, and automated decision-making, raising questions about the true openness of the model’s use and licensing terms. The release has sparked discussion about the difference between open weights and open source, especially given the absence of the full training data or pipeline.
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 for AI Development
The release of Inkling under open licensing is a significant development because it challenges the prevailing industry norm of closed, proprietary models. By making the full weights publicly available, Thinking Machines enables a broader range of organizations and researchers to experiment, fine-tune, and deploy the model independently, potentially accelerating innovation and transparency in AI.
However, the existence of a separate Acceptable Use Policy and the lack of transparency around training data and pipeline raise questions about how ‘open’ the model truly is. This development could influence future AI licensing practices and set a precedent for balancing openness with responsible use, especially in sensitive domains like security and public safety.
Vision-Language Models in Production: Architecting Multimodal LLM Applications: From Vision-Language API to Self-Hosted Model (Production AI Engineering Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Recent Trends in AI Model Transparency and Open-Source Releases
Over the past year, the AI community has seen increased calls for transparency, with some organizations releasing models with open weights and others withholding full training data or pipelines. The recent release of Meta’s Llama 2 and the open-sourcing of models like Stable Diffusion exemplify this trend. However, debates continue about the scope of openness, licensing restrictions, and responsible use policies layered on top of open weights.
Thinking Machines’ approach with Inkling is notable because it offers full weights immediately, coupled with a candid acknowledgment that it is not the most powerful model, and transparency about training methods and benchmarks. This contrasts with many proprietary models that emphasize performance over openness, and it reflects a broader shift toward democratizing access to advanced AI tools.
“We believe in providing the community with the tools to innovate responsibly and transparently.”
— Thinking Machines spokesperson
Unresolved Questions About Inkling’s Openness and Use Restrictions
It remains unclear how enforceable the separate Acceptable Use Policy is, and whether it significantly limits the model’s open-source potential. Details about the training data, pipeline, and whether the policy applies to all derivatives of the model are not fully verified. Additionally, the long-term impact of this release on industry standards for openness and licensing is still developing.
Next Steps for AI Community and Industry Adoption
Expect further analysis and independent benchmarking of Inkling’s performance and safety features. Organizations will likely review the licensing and use policies carefully before adopting or modifying the model. Additionally, other AI developers may follow suit, balancing openness with responsible use, and potentially prompting new standards for licensing and transparency in the industry.
Key Questions
What makes Inkling different from other large language models?
Inkling is notable for its open weights under Apache 2.0, its multimodal capabilities supporting text, images, and audio, and its transparent training methodology, although it is not the most powerful model currently available.
Does the open release mean anyone can freely use and modify Inkling?
While the weights are openly available under Apache 2.0, the presence of a separate Acceptable Use Policy and lack of detailed training data means users should review licensing terms carefully before deploying or modifying the model.
Why is the release of Inkling considered significant?
It represents a shift toward greater transparency and democratization in AI, challenging the norm of proprietary models, and could influence future licensing and development practices across the industry.
What are the main concerns about Inkling’s openness?
The main concerns involve the potential restrictions imposed by the Acceptable Use Policy, the lack of transparency about training data, and whether the open weights alone constitute genuine open-source access.
What should we expect next from Thinking Machines?
The company will likely publish further benchmark results, release the smaller Inkling-Small model, and clarify licensing and use policies. Industry reaction and adoption will also be closely watched.
Source: ThorstenMeyerAI.com
