📊 Full opportunity report: Understanding The Underlying Signals In Thinking Machines’ Inkling on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Thinking Machines has publicly released the full weights of its new multimodal model, Inkling, under an open license. However, the company maintains a separate use policy, raising questions about true openness and control. This move highlights ongoing debates about open-source AI models and ownership costs.
Thinking Machines has released the full weights of its latest multimodal model, Inkling, under an Apache 2.0 license, making it openly accessible for download, modification, and deployment. This marks a significant departure from typical proprietary or restricted model releases, emphasizing transparency and user ownership.
The Inkling model is a 975-billion-parameter mixture-of-experts transformer supporting multimodal input—text, images, and audio—processed jointly within a shared space. It was trained on 45 trillion tokens across various media types, with a 1-million-token context window. The model’s weights were made available on Hugging Face under Apache 2.0, allowing broad use and modification.
Despite this open release, the company states that a separate Model Acceptable Use Policy (AUP) restricts certain applications, such as surveillance, deception, and automated decision-making affecting individuals’ rights. The AUP’s scope and enforceability are not fully verified, raising questions about the true openness of the model. The company also clarified that the open weights do not include the training data or full training pipeline, which remain proprietary.
Additionally, the model achieved competitive scores on various benchmarks, notably excelling in safety metrics like refusal of violence requests, and performed well in speech and multimodal tasks. However, it remains mid-tier on some language benchmarks, highlighting its varied strengths and weaknesses.
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 Weights and Usage Restrictions
This release underscores a shift toward more transparent AI development, giving users direct access to model weights for customization and deployment. However, the accompanying use restrictions complicate the notion of true open source, as they may limit certain applications and raise legal questions. For developers and organizations, understanding these distinctions is critical for compliance and strategic planning. The move also prompts broader industry debate about what constitutes genuine openness versus controlled access.
USB-C cables for data transfer
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Industry Norms and the Rise of Open-Weight Models
Historically, most large models have been released as closed-source, with access controlled via APIs or licensing. Recent years have seen a handful of open-weight releases, often accompanied by proprietary training data and restrictions. Thinking Machines’ approach—releasing full weights under Apache 2.0—aligns with a growing trend of transparency, but the addition of a restrictive AUP reflects ongoing tension between openness and control. The timing coincides with increased scrutiny of AI safety, ownership costs, and the ethics of open models.
Prior to Inkling, notable open-weight models include LLaMA and Falcon, which also faced similar debates about licensing and restrictions. The current development indicates a possible shift toward more openly accessible models, but with caveats that could influence how developers and companies adopt them.
“Our goal is transparency and user ownership. The AUP is about responsible use, not restricting access to the model itself.”
— Thinking Machines spokesperson
4K smart TVs for sports
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About Model Licensing and Use
It is not yet clear how strictly the AUP will be enforced in practice, or how it might impact commercial or research use. The precise scope of restrictions, especially regarding surveillance or automated decision-making, remains unverified. Additionally, the implications of not releasing training data or full training pipeline are still being assessed by the community.
remote office setup accessories
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for Adoption and Verification
Expect further analysis of Inkling’s licensing terms and real-world testing of its capabilities. Organizations will likely scrutinize the AUP and seek legal clarification before deploying. The company may also release additional documentation or updates to clarify enforcement measures. Meanwhile, the broader AI community will monitor whether this approach influences future model releases and licensing strategies.
Generative AI for Software Development: Building Software Faster and More Effectively
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What does releasing the full weights under Apache 2.0 mean?
It allows anyone to download, modify, and deploy the model freely, as long as they comply with the license and any additional policies.
Are there restrictions on how I can use Inkling?
Yes, according to reports, Thinking Machines maintains a separate use policy that restricts applications like surveillance, deception, and automated decision-making affecting individuals’ rights. The enforceability of these restrictions is not fully verified.
Does open-weight release mean the model is fully open source?
No, the weights are under Apache 2.0, but the training data, pipeline, and certain use restrictions are not openly shared, which limits the scope of true open source.
Why is this release significant for AI development?
It demonstrates a move toward greater transparency and user ownership of large models, but also highlights ongoing debates about openness versus control in AI licensing.
Source: ThorstenMeyerAI.com
