Inkling is Thinking Machines' first open-weights multimodal LLM. Honest comparison vs Kompozy: when an open model you customize wins, and when you need a content engine.
If you landed here comparing "Inkling vs Kompozy," the first thing to say is that they are not the same kind of thing. Inkling is a foundation model — open weights you download, host, and fine-tune. Kompozy is a content engine you log into and use. One is an engine block; the other is the finished car. So this is less "which is better" and more "which layer of the stack are you actually shopping for."
I run Kompozy, and I am not going to pretend Inkling is a competitor we out-feature — it does a different job, and it does it impressively. Inkling is Thinking Machines Lab's first model, from the team led by former OpenAI CTO Mira Murati, released July 15, 2026 as open weights under Apache 2.0. It is a large, natively multimodal mixture-of-experts model (975B total, ~41B active) that reads text, images, and audio and reasons in text, with controllable thinking effort and calibrated, uncertainty-aware answers. At launch, Artificial Analysis rated it the leading U.S. open-weights release on its Intelligence Index. If your reason for looking is "I want a frontier-grade open model I can run, fine-tune, and specialize myself," Inkling is one of the strongest answers on the market and Kompozy is not what you want.
The split is simple. Inkling gives you a reasoning-and-text model — and nothing downstream of text. It generates no images, no video, no captions on a clip, and it publishes nothing, because that was never its goal. Its whole pitch is a bet against one-size-fits-all AI: an open base you customize rather than a closed system you rent. To turn Inkling into published content you would build the entire production and distribution pipeline yourself — inference hosting, media generation, brand styling, a scheduler, and platform integrations. Kompozy is that pipeline, already built, running on managed Claude and OpenAI models. If your bottleneck is shipping content, not operating a model, that is the real comparison.
Everything below reconciles Inkling against Thinking Machines' own release and the Hugging Face model card, and Kompozy pricing against ours, both checked on 2026-07-17.
Inkling is the first in-house model from Thinking Machines Lab, released on July 15, 2026 as open weights under the permissive Apache 2.0 license. It is a natively multimodal mixture-of-experts model — 975 billion total parameters with about 41 billion active per token — that accepts text, image, and audio input and produces text output, with up to a 1M-token context on the open weights (256K via the Tinker fine-tuning API) and training on roughly 45 trillion tokens. Two design choices define it: controllable "thinking effort," so you can trade reasoning depth for speed and cost, and calibration — it is built to flag uncertainty rather than guess. It is positioned as an open base to be fine-tuned and specialized, not a single general-purpose system you rent, and at launch Artificial Analysis rated it the leading U.S. open-weights model on its Intelligence Index while noting strong token efficiency. A lighter preview, Inkling-Small (276B/12B), targets low-latency workloads. What it does, concretely, is generate and reason over text — drafts, summaries, analysis, answers — and it can reason over images and audio you feed it. What it does not do is anything downstream of text. There is no image or video generation, no captioning, no design templates, no scheduler, and no platform publishing. Those are jobs for the application layer you build on top of it. You reach Inkling by downloading the weights from Hugging Face, fine-tuning on Tinker, or using hosted inference through partners like Together AI, Fireworks, Modal, Databricks, and Baseten.
The reason to look past "just use Inkling" for a content workflow is that a raw model is a long way from a published post. To go from Inkling to a TikTok or a LinkedIn carousel you would need to host inference, wire up image and video generation (Inkling does neither), build brand-styling and caption rendering, write a scheduler, and integrate every platform API — real engineering before a single post ships, plus the GPU bill to serve a 975B-parameter model at any throughput. That is the right investment for a lab, a product team building on an open base, or a company whose core requirement is control over the model itself. It is the wrong investment for a creator or agency whose job is to publish. None of this is a knock on Inkling. It is doing exactly what it set out to do — be a strong, open, customizable base layer, deliberately the opposite of a locked general-purpose product. It just sits one or two layers below the problem most content creators have. If you want the openness, the fine-tuning, and the self-hosting, Inkling is excellent and you should use it. If you want finished, on-brand, scheduled content across platforms, you want the layer on top — and you probably do not want to build that layer yourself.
| Feature | Inkling | Kompozy | Note |
|---|---|---|---|
| Open weights (Apache 2.0) | Yes | No | Inkling ships downloadable weights for commercial use. Kompozy is a managed product, not an open model. |
| Self-host / fine-tune on your own infrastructure | Yes | No | Inkling is downloadable and fine-tunable via Tinker. Kompozy is hosted SaaS. |
| Native multimodal input (text, image, audio) | Yes | Partial | Inkling reasons over images and audio you feed it. Kompozy ingests source media but is not a general reasoning model. |
| Controllable thinking effort / cost tuning | Yes | N/A | Inkling lets you dial reasoning up or down. That is a model-inference feature, not a content-engine one. |
| Calibrated, uncertainty-aware answers | Yes | N/A | A genuine Inkling strength for trustworthy reasoning. Kompozy is a generation/publishing layer, not a reasoner. |
| AI text generation (captions, scripts, blogs) | Partial | Yes | Inkling generates raw text. Kompozy writes on-brand copy governed by a Persona Brief. |
| AI image generation | No | Yes | Inkling outputs text only. Kompozy renders photo posts, carousels, quote cards, infographics. |
| AI / avatar video generation | No | Yes | Inkling produces no media. Kompozy ships persona/avatar video, clips, marketing shorts. |
| Branded captions + design templates (HyperFrames) | No | Yes | No design layer in a raw model. Kompozy renders pixel-exact brand styling. |
| Scheduling + autopilot | No | Yes | Inkling has no scheduler. Kompozy ships a calendar, autopilot, and review pipeline. |
| Multi-platform publishing (9 platforms + email + blog) | No | Yes | Inkling publishes nothing. Kompozy fans output to all destinations from one queue. |
| Works without ML engineering / GPUs | No | Yes | Running Inkling well needs infra and ops. Kompozy is log-in-and-use. |
| Tier | Inkling plan | Inkling price | Kompozy plan | Kompozy price |
|---|---|---|---|---|
| Entry | Inkling (self-hosted) | Free weights (Apache 2.0) + your own GPU/inference cost | Kompozy Creator | $49/mo (2,500 credits) |
| Mid | Inkling via partner inference / Tinker | Provider inference pricing (Tinker fine-tuning ran a launch discount) | Kompozy Pro | $299/mo (18,000 credits) |
| Top | Inkling fine-tuned / on-prem | Engineering + infra (custom) | Kompozy Enterprise | Custom (sales-led) |
Here is the honest pitch, because Inkling and Kompozy live on different floors of the same building. Inkling is a foundation model — and a strong one, because it is open under Apache 2.0, natively multimodal, calibrated, and rated the top U.S. open-weights release at launch. If your problem is "I want a frontier model I can run, audit, and fine-tune myself," Inkling is a genuinely great answer and you should not be reading a Kompozy page for it.
But a model is not a content operation, and Inkling was deliberately built as a base to customize — not a finished product. To get from Inkling to a published TikTok, Reel, carousel, or newsletter you would build everything that sits above the model: inference hosting for a 975B-parameter system, image and video generation (Inkling does neither), brand styling and captions, a scheduler, and integrations for nine platforms. That is a serious engineering project plus a GPU bill — and even fully fine-tuned, Inkling still only outputs text. Kompozy is that entire layer, already built and managed: it generates 18 content formats across video, image, text, blog, and newsletter, holds one brand voice through a Persona Brief, and publishes to nine platforms plus email and blog on a schedule, on autopilot.
The cleanest way to think about it: if you care most about owning and customizing the model, choose Inkling. If you care most about producing and shipping content, choose Kompozy — and if you want both, you can draft in your own fine-tuned Inkling deployment and let Kompozy turn those drafts into finished, scheduled posts. Start on Kompozy Creator at $49/mo (2,500 credits) to test the production half.
Not really — they sit at different layers. Inkling is an open foundation model you download, host, and fine-tune; Kompozy is a content generation and publishing engine you log into. People compare them because both involve AI, but Inkling produces raw text while Kompozy produces finished, scheduled posts across platforms. For most content workflows they are complementary, not competing.
Inkling can draft and reason over text — and it can read a recording or screenshot you feed it — but it cannot create images or video, design posts, or publish anything. To turn an Inkling draft into published content you either build that pipeline yourself or use a content engine like Kompozy that generates the media and publishes to nine platforms.
When you want to run, fine-tune, or control a frontier-grade open model yourself — a product team building on an open base, or an organization that needs the model and its inference on its own infrastructure. In those cases an open model like Inkling is exactly right and a hosted content SaaS is not.
Inkling itself is free under Apache 2.0 — but your real cost is the GPU/inference hosting for a 975B-parameter model plus the pipeline you build around it, or a partner provider's inference pricing (Tinker fine-tuning ran a launch discount). Kompozy is a managed subscription starting at $49/mo (2,500 credits) for Creator and $299/mo (18,000 credits) for Pro, with no infrastructure to run.
Yes. Draft and reason in your own fine-tuned Inkling deployment — including turning a recording or image into structured text — then bring those drafts into Kompozy to generate the video, images, and carousels and publish across platforms. Inkling owns the open, customizable text; Kompozy owns the media and the publish.