Inkling AI API Access & Pricing

From free weights to enterprise gateways — every channel to use Inkling AI, what each costs, and which fits your use case.

Every Way to Access Inkling AI

Unlike OpenAI-style models with one API and one price sheet, Inkling AI reaches you through five distinct channels — from completely free weights to enterprise gateways:

Channel Cost What You Get
Hugging Face (weights) Free (Apache 2.0) Full BF16 weights + official NVFP4 variant. You handle hosting and inference costs yourself.
Tinker Playground Free to try Browser chat interface in the Tinker console — the fastest way to feel out the model before committing.
Tinker API + fine-tuning Usage-based Managed inference and fine-tuning. The launch demo's full self-finetune run reported $0.00 spent during the promo period.
Databricks Unity AI Gateway Databricks pricing Enterprise route — Inkling as a governed endpoint next to your data, for building agents and applications.
Unsloth GGUFs (self-host) Free download Community dynamic quants from 270GB for local inference with llama.cpp — see our how-to-run guide.

Which Channel Should You Use?

Just curious

Tinker Playground — free browser chat, zero setup. Five minutes from reading this to talking with Inkling AI.

Building an app prototype

Tinker API — usage-based billing with no infrastructure. Move to self-hosting only when volume justifies it.

Enterprise with data governance

Databricks Unity AI Gateway — Inkling next to your lakehouse with access controls your compliance team already approved.

Sustained heavy inference

Self-host quantized weights on rented GPUs (~$20–30/hr for 8×H100) — per-token pricing loses to dedicated compute at scale.

Expect third-party hosts (Together, Fireworks, OpenRouter and the like) to add per-token Inkling AI endpoints soon — that is what happened within weeks for DeepSeek and GLM. We will update this page as they land.

API & Pricing FAQ

Is the Inkling AI API free?
The weights are free (Apache 2.0), and the Tinker Playground is free to try. API inference and fine-tuning on Tinker are usage-based; Databricks access follows Databricks pricing. Self-hosting is 'free' minus your very real GPU costs.
What is the cheapest way to use Inkling AI?
For casual testing: the free Tinker Playground. For light API use: Tinker's usage-based billing. For heavy sustained inference: self-hosting quantized weights on rented GPUs typically beats per-token pricing at scale.
Does Inkling AI have per-token API pricing like OpenAI?
Thinking Machines has not published a public per-token price sheet — access runs through Tinker (usage-based) and partners like Databricks. Third-party hosts will likely offer per-token Inkling endpoints soon, as they did for DeepSeek and GLM.
How much does it cost to self-host Inkling?
Roughly $20–30/hour for an 8×H100 node that comfortably serves quantized Inkling, or a one-time ~$10K+ Mac Studio Ultra for slow-but-local 1-bit inference. Our how-to-run guide breaks down every tier.

Go Deeper

Self-Hosting Guide Fine-Tuning on Tinker