# Giotto Alto and Solo: API Pricing and Private Deployment Explained

The Swiss lab is offering reasoning models through cloud access and customer-controlled infrastructure, with separate API billing options to understand.

By Paras Madan | 2026-10-29


Giotto announced Alto and Solo on October 27, a reasoning-model family offered through European cloud services and private deployments. Alto targets demanding workloads, while Solo is designed for a single GPU. The practical question for buyers is which deployment and billing arrangement fits their workload, not simply where the company is based.

## What Giotto Alto and Solo offer

The Lausanne company's [launch announcement](https://www.webdisclosure.com/press-release/giottoai-sa-etr-giotto-launches-sovereign-ai-family-to-make-ai-public-infrastructure-BwePS9mOKXG) says Alto runs on eight B300 GPUs and Solo on one GPU. These are vendor deployment claims, not performance results from independent testing.

Giotto also announced free shared-cloud access, anonymous chat and mobile apps. Private infrastructure is a different commercial arrangement: the release describes monthly dedicated-cloud rentals and annual or multiyear on-premises licences priced by GPU count and type.

A single-GPU deployment could be useful for a team that wants to keep a narrowly defined workflow inside its own environment. But a GPU count alone does not establish affordability or speed. The hardware specification, concurrent users and context length still determine whether it is a practical fit.

## Giotto Alto API token pricing

Giotto's [public API page](https://www.giotto.ai/api) currently displays these euro prices for Alto:

| Token category | Price per million tokens |
| --- | --- |
| Input | €2.10 |
| Output | €10.50 |
| Cache hit | €0.21 |
| Cache write | €2.10 |

At those displayed rates, one million ordinary input tokens plus 100,000 output tokens would cost €3.15, before any other applicable charges. That is arithmetic for a defined usage mix, not a prediction of the cost of an agent or document workflow.

Repeated inputs and long generated answers produce different bills. The cache-hit price makes reuse relevant, while the output rate means a long reasoning response should not be budgeted as if all tokens cost the input rate.

## Flat subscription and by-token access are separate

The launch release also describes a flat-fee API option with one concurrent request and token packages for greater concurrency. Reading that alongside the pricing table can look inconsistent until the product documentation is considered.

Giotto's [API setup guide](https://giottoai.zendesk.com/hc/en-001/articles/37073730796818-API), updated October 28, distinguishes a flat subscription from a separate By-token API product. It describes separate credentials and €10 credit packs. The public token table therefore should not be treated as the price of the flat subscription, whose fee is not stated on that page.

The guide documents a Python client with a Responses interface and medium or high reasoning effort. It also distinguishes orchestrator-event streaming from token-by-token text streaming. Developers building a live interface should check that difference before assuming a familiar streaming experience.

## Private deployment is not an open-model licence

The [deployment offering](https://www.giotto.ai/offering) includes managed European hosting, installation on private GPUs and preconfigured hardware. Its private-deployment and hardware links lead to sales enquiries, so the site should not be read as an unrestricted public weights download.

There is a licensing distinction too. Giotto's [individual EULA](https://www.giotto.ai/eula) describes a proprietary model and limits that licence to non-commercial testing and evaluation. Business buyers should confirm the agreement attached to their selected product rather than assume free chat grants production rights.

**Control over where a model runs and freedom to modify or redistribute it are different things.** A private deployment can address an infrastructure requirement without removing reliance on a vendor's software, updates or contract.

For an enterprise evaluation, the useful next step is a small, representative workload with agreed hardware, latency and licensing requirements. Giotto's deployment choices make it a candidate for that discussion; they do not replace the discussion with a sovereignty label.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical teams explain complex AI products and deployment tradeoffs in language developers and decision-makers can use.

[Book a 30-minute call](https://cal.com/paras-madan/30min).