TL;DR
Thinking Machines Lab released its first foundation model, Inkling, with downloadable weights under Apache 2.0 and immediate support from major inference tools. The launch puts model ownership and deployment flexibility ahead of benchmark leadership, although hardware costs, license questions and unverified performance claims limit the practical impact.
Thinking Machines Lab, founded by former OpenAI technology chief Mira Murati, released its first foundation model, Inkling, on July 15 with full downloadable weights under Apache 2.0 and immediate support from leading inference frameworks. The release matters less as a claim to benchmark leadership than as a decision to give developers model control from day one.
According to the lab’s launch materials and Hugging Face repository, Inkling is a 975-billion-parameter Mixture-of-Experts model that activates 41 billion parameters for each token. It supports a one-million-token context window and accepts text, images and audio, producing text responses. The lab says it trained the model on 45 trillion multimodal tokens.
The release includes BF16 and NVFP4 checkpoints, with day-zero support in transformers, vLLM, SGLang and llama.cpp. Apache 2.0 generally permits downloading, modification and commercial use, giving organizations more control than a hosted API alone. However, the training data and full training pipeline were not published, meaning Inkling is an open-weight release rather than a fully open-source training project.
Thinking Machines also acknowledged that Inkling is not the strongest available model, whether compared with closed or open systems. Vendor-published results put it ahead on some mathematics, scientific reasoning, audio and adversarial tests, but behind rivals including GLM-5.2 and Kimi K2.6 on several coding, agentic and multimodal evaluations. Those results have not yet received broad independent replication.
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.
Model Ownership Gains Strategic Weight
The release gives companies, researchers and public institutions a Western-developed open-weight option that can be hosted, modified and retained without relying entirely on a provider’s API. That can reduce exposure to service changes, access restrictions and vendor pricing, while allowing private deployment for sensitive workloads.
Inkling also includes an adjustable reasoning-effort setting from 0.2 to 0.99, allowing operators to trade response quality against token use, latency and cost. Thinking Machines reportedly found that the model matched Nemotron 3 Ultra on Terminal-Bench 2.1 while using about one-third of the tokens. If independently confirmed, that cost curve could matter more to high-volume deployments than a model’s highest benchmark score.
AI model deployment tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Deployment-First Open Release
Open weights have often arrived after a commercial API or as a smaller companion to a closed flagship. Thinking Machines reversed that sequence by publishing complete flagship checkpoints first, pairing them with an established license and broad inference support. The choice aligns the launch with buyers seeking sovereignty and operational control, rather than only access to another hosted chatbot.
The lab is about 17 months old and includes former OpenAI personnel who worked on ChatGPT. It also previewed Inkling-Small, a 276-billion-parameter model with 12 billion active parameters. The lab says the smaller version matches or exceeds the flagship on several tests, but its full weights are still awaiting release.
“Inkling is not the strongest model available today, closed or open.”
— Thinking Machines Lab’s launch announcement
open-source AI model weights
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License Limits and Benchmark Gaps
It is not yet clear whether a separate Model Acceptable Use Policy adds restrictions covering surveillance, deception or automated decisions affecting legal rights. Thorsten Meyer AI reported such a policy but said it had not independently verified the language. Organizations in geospatial intelligence, public safety or other regulated fields will need to inspect the repository’s current terms before deployment.
Performance claims also remain provisional. Several scores are vendor-published, some may reflect a prerelease checkpoint, and independent testers have not yet reproduced the full results. Practical access is another limitation: the analysis estimates that BF16 deployment needs at least two terabytes of aggregate VRAM, while NVFP4 still requires roughly 600 gigabytes.
multimodal AI development kit
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Tests and Smaller Weights
Researchers and prospective users will now test Inkling on real workloads, compare its cost and accuracy with GLM-5.2, Kimi K2.6 and closed APIs, and verify the applicable use restrictions. The next major milestone is the release of Inkling-Small’s full weights, which may offer a more practical route for organizations unable to host the flagship. Independent benchmark replication and measured deployment costs will determine whether the launch’s ownership promise produces a competitive operating advantage.
AI inference frameworks
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Inkling?
Inkling is Thinking Machines Lab’s first multimodal foundation model. It uses a Mixture-of-Experts architecture with 975 billion total parameters and 41 billion active parameters per token.
Are Inkling’s weights available for commercial use?
The published weights carry an Apache 2.0 license, which generally permits commercial use and modification. Users should also check whether a separate acceptable-use policy applies to their planned deployment.
Is Inkling the highest-performing open model?
No. Thinking Machines says Inkling is not the strongest model available. Its reported results are competitive in several categories, but GLM-5.2 and Kimi K2.6 lead on some coding, reasoning and multimodal tests.
Can Inkling run on a local workstation?
The flagship is beyond most workstations. Reported estimates call for at least 600 gigabytes of memory using NVFP4 and about two terabytes for BF16, making multi-accelerator infrastructure the likely requirement.
Why are the open weights the main story?
They let organizations host, modify and retain the model rather than depend only on a remote API. That changes the discussion from temporary access to ownership, deployment control and long-term availability.
Source: Thorsten Meyer AI
