Does Inkling NVFP4 support zero data retention?

Yes. Inkling NVFP4 is available with zero data retention on Modal through the Opper gateway.

Where is Inkling NVFP4 hosted?

See the per-route table for Inkling NVFP4 hosting locations and residency details.

Is Inkling NVFP4 trained on prompts and outputs?

No. Every hosting route for Inkling NVFP4 is configured so customer prompts and outputs are not used to train models by default.

Is a GDPR DPA available for Inkling NVFP4?

Yes. A GDPR Data Processing Agreement is available on at least one route serving Inkling NVFP4. Contact hello@opper.ai to receive it.

Inkling NVFP4

by Thinking Machines

Inkling NVFP4 is the 4-bit NVFP4 quantization of Inkling, the from-scratch model Thinking Machines Lab published on July 15, 2026 under Apache 2.0. Underneath is a 66-layer decoder-only Mixture-of-Experts transformer with 975 billion total and 41 billion active parameters, where each token is handled by 6 routed experts out of 256 alongside 2 shared ones. Pretraining covered 45 trillion tokens spanning text, image, audio, and video, and the model takes text, images, and 16 kHz audio (ideally under 20 minutes) as input, generating text across a context window of up to 1 million tokens. Thinking effort is adjustable, and the lab's efficiency pitch is that Inkling reaches a given score with fewer output tokens than comparable models, for instance landing level with Nemotron 3 Ultra on Terminal-Bench 2.1 while spending about a third as many tokens. Published results include 97.1% on AIME 2026, 87.2% on GPQA Diamond, 77.6% on SWE-bench Verified, 63.8% on Terminal-Bench 2.1, 73.5% on MMMU Pro, and 91.4% on VoiceBench. Thinking Machines does not claim the top spot for Inkling and instead pitches it as an open base to customize, with fine-tuning on its Tinker platform available from day one. The NVFP4 checkpoint brings the aggregate memory requirement down from at least 2 TB in BF16 to about 600 GB, enough to run on 4 B300 or 8 H200 GPUs, which is what makes a model this size deployable on a single eight-GPU system.

Call Inkling NVFP4 on Opper with the OpenAI SDK. Sign up without a credit card. Get started

Get your API key Try in chat Compare

Key info

Input

Output

Features

Context window

Max output

262K

Input price

$1.20 /1M$0.27 cache read

Output price

$5.00 /1M

Zero data retention on pay-as-you-go
No training by default
GDPR DPA available

Available routes

Inkling NVFP4 runs on 1 route through the Opper gateway. Compare residency, zero data retention and training posture at a glance, with full data-handling detail per route below.

Provider	Region	Zero data retention	Training	Input	Output	Cache read
Modal	Multi		No (opt-in)	$1.20	$5.00	$0.27

Data handling per route

Each route hosting Inkling NVFP4 has its own privacy posture, residency, and GDPR terms. Postures are maintained by Opper with a last-verification timestamp.

Get started

Call Inkling NVFP4 through the Opper gateway with one API key. Opper is drop-in compatible with the OpenAI, Anthropic and Google AI SDKs.

Inkling NVFP4 is a premium model on Opper. Sign up needs no credit card: you get an API key straight away and the free models work in the playground and the API. Add a card to use premium models, pay-as-you-go with no minimum.

Set it up with your agent

Copy this and paste it into a coding agent like Claude Code, Cursor or Codex and it'll wire up Opper for you.

Or call it directly

1import OpenAI from "openai";
2
3const client = new OpenAI({
4  apiKey: process.env.OPPER_API_KEY,
5  baseURL: "https://api.opper.ai/v3/compat",
6});
7
8const completion = await client.chat.completions.create({
9  model: "modal/inkling-nvfp4",
10  messages: [{ role: "user", content: "Hello" }],
11});
12console.log(completion.choices[0].message.content);