---
type: analysis
slug: research-giotto-has-not-documented-alto-s-lineage-and-the-reported-kimi-admission-is
title: "Giotto has not documented Alto's lineage, and the reported Kimi admission is unverified"
description: "Giotto markets Alto and Solo as sovereign European models, but it publishes no base model, weights or parameter count. The claim that Alto inherits Kimi weights rests on user screenshots and a reported CEO remark that we could not trace to its source."
published_at: 2026-11-13T16:00:00.000Z
topics: ["reasoning","research"]
source_urls: ["https://www.giotto.ai/","https://www.eqs-news.com/news/corporate/giotto-launches-sovereign-ai-family-to-make-ai-public-infrastructure/165f1e84-de97-428b-b77a-9e3c65dd5b57_en","https://explainx.ai/blog/giotto-alto-solo-swiss-sovereign-ai-models-benchmarks-claimed-vs-verified-2026","https://pasqualepillitteri.it/en/news/27761/giotto-ai-alto-solo-sovereign-european-ai-models","https://tensoria.fr/blog/giotto-alto-solo-ia-souveraine-europe","https://digg.com/ai/sqxq346m","https://www.giotto.ai/eula","https://venturebeat.com/technology/kimi-k3s-full-weights-are-here-but-theyre-open-with-a-caveat-what-enterprises-should-know","https://kiwoche.com/1396/giotto-ai-schweizer-reasoning-modelle-solo-alto-souveraene-ki-kimi/"]
author: "Andrey Reshetnikov"
author_profile: https://newruntime.com/owner-profile.md
---
On 27 October 2026, Lausanne-based Giotto launched two reasoning models. Alto is its flagship and runs on eight Nvidia B300 GPUs; Solo runs on a single GPU. Giotto sells both as sovereign European AI, hosted in Europe or on the customer's own hardware. Within hours, users posted screenshots of Alto calling itself a Kimi model from Moonshot AI, served under the name "giotto-model". Critics then said Alto is a post-trained version of Moonshot's open-weight Kimi K3.

This matters to builders because sovereignty claims are about who controls the weights, the license and the supply chain, not just where the servers sit. If Alto is a Kimi K3 derivative, buyers inherit Moonshot's license terms and the questions about model origin that their procurement teams care about. Giotto has neither confirmed nor documented this.

## What we checked

- **Can Giotto independently document which model weights Alto and Solo inherit, and what training or test-time changes distinguish them from Kimi?** refuted. Giotto has published no such documentation. Its homepage, its 27 October press release and its EULA name no base model, weights, architecture or parameter count. Independent reviewers found no open weights or training-data statement, so nothing outside the company can confirm or rule out a Kimi lineage. [1][2][3][4]
- **Did CEO Aldo Podestà publicly confirm that Alto inherits Kimi weights?** unresolved. One German outlet, KI Woche, is said to report that Podestà confirmed on X that Alto took over Kimi weights and that Giotto added its own reasoning and fine-tuning on top. The version of that article we opened does not contain the statement, his X profile could not be retrieved, and reports from 28 October say Giotto had not commented. [9][5][4]
- **Has Giotto changed its launch EULA's restrictions on benchmarking and reverse engineering?** partly confirmed. At launch, the EULA as quoted by Tensoria barred publishing benchmarks without Giotto's prior written consent. The EULA now online, dated October 2026, still bans reverse engineering but has no benchmarking clause, which points to a quiet revision. Giotto has not announced or explained any change. [5][7][6]

## What we found

- Giotto's own materials say a lot about where the models run and nothing about where they came from. The homepage promises deployment on European cloud or private infrastructure. The press release describes Alto as the flagship for reasoning workloads but discloses no architecture, training method or base model. On the homepage, Kimi K3 appears only as a competitor in Alto's benchmark table. [1][2]
- Independent reviewers found nothing to check the lineage against. ExplainX found no open weights and no parameter count, and found no third-party reproduction of Giotto's benchmark numbers. An Italian report from 28 October said Giotto had not named Alto's base model or responded to the Kimi accusation. [3][4]
- The launch EULA restricted inspection of the model. Tensoria quoted a clause barring published benchmarks without Giotto's consent, and Digg reported that the license allowed only noncommercial testing and evaluation. The EULA now online still bans reverse engineering and extraction. It refers to open-source components under their own licenses but names no model. It no longer includes the benchmarking clause. [5][6][7]
- Kimi K3 is a plausible base, and its license would matter to buyers. Moonshot released the weights on 27 July 2026: a 2.8-trillion-parameter mixture-of-experts model under a custom license. That license requires large commercial users to display 'Kimi K3' in their product and requires large model-as-a-service operators to sign a separate agreement. Sizing Alto for eight B300s fits a model of that scale, but that is circumstantial evidence, not proof. [8]

## What it means

Treat Alto and Solo as models of undisclosed origin. Hosting in Europe answers where your data goes. It does not answer whose weights you are running, which license terms you inherit, or what was changed in post-training. If you are evaluating Giotto for regulated or public-sector work, ask for the base model and its version, the upstream license, a post-training summary, and written permission to run and publish your own evaluations, before you accept any sovereignty claim.

Do not rely on the model's own self-identification either way. A model calling itself Kimi is strong circumstantial evidence, but distillation from Kimi outputs could produce the same answer. Giotto's ban on reverse engineering also stops customers from checking for themselves. If Giotto confirms a Kimi K3 lineage, check whether your deployment crosses the K3 license thresholds for attribution or a separate model-as-a-service agreement.

## Competing readings

- **Alto is a post-trained version of Kimi K3 weights with Giotto's own reasoning and fine-tuning on top** (likely, about 65%). What would tell it apart: Users saw Alto identify as Kimi, Giotto has not denied it, and Alto's eight-B300 footprint fits a model of K3's scale [3][4][7]; Giotto's EULA bans reverse engineering, which blocks outside checks on the weights [6].
- **Alto is Giotto's own or a differently sourced model that learned to call itself Kimi from distilled or synthetic Kimi outputs** (unlikely, about 25%). What would tell it apart: Self-identification does not prove where weights came from, and no primary statement of lineage exists [3][4].
- **Alto is mostly a hosted wrapper or router around another provider's model rather than a separately trained derivative** (very unlikely, about 10%). What would tell it apart: Giotto offers on-premises installs on its own GPU racks, which points against a pure API wrapper [1][2].

## Base rate

Of 81 earlier announcement stories about companies or models in the New Runtime archive over the last 12 months that are at least 3 months old, 80 (99%) were followed within 3 months by a release or access-change story on the same company or product line. In 80 of 81 earlier announcement stories, a release or access-change story on the same company followed within 90 days. That makes further Giotto product news very likely, but it says nothing about whether Giotto will disclose Alto's lineage.

## What would change this view

- Giotto publishes a model card or open-source notice naming a non-Kimi base model, or releases weights that can be fingerprinted (model-listing signal): The Kimi-derivative reading would be weakened or refuted.
- Giotto's open-source documentation or user interface adds a Kimi K3 attribution (signal): This would confirm the Kimi lineage and that Giotto is complying with Moonshot's license.
- Independent fingerprinting or logit-similarity tests comparing Alto with Kimi K3 are published (signal): This would settle the lineage question either way without needing Giotto's cooperation.

## Still open

- Did Aldo Podestà actually say on X that Alto inherits Kimi weights, and in what words?
- If Alto is built on Kimi, which version is it, and does Solo share that lineage or come from a different, smaller base?
- What post-training or test-time methods does Giotto add, and how much do they change benchmark results?
- When and why was the benchmarking clause dropped from the EULA, and does Giotto's documentation list Kimi among its open-source components?

## Sources

1. [Giotto.ai homepage](https://www.giotto.ai/) (primary)
2. [Giotto launches sovereign AI family (press release via EQS)](https://www.eqs-news.com/news/corporate/giotto-launches-sovereign-ai-family-to-make-ai-public-infrastructure/165f1e84-de97-428b-b77a-9e3c65dd5b57_en)
3. [Giotto Alto and Solo: claimed vs verified](https://explainx.ai/blog/giotto-alto-solo-swiss-sovereign-ai-models-benchmarks-claimed-vs-verified-2026)
4. [Giotto AI launches Alto and Solo](https://pasqualepillitteri.it/en/news/27761/giotto-ai-alto-solo-sovereign-european-ai-models)
5. [Tensoria on Giotto Alto and Solo](https://tensoria.fr/blog/giotto-alto-solo-ia-souveraine-europe)
6. [Giotto's AI license bans reverse engineering](https://digg.com/ai/sqxq346m)
7. [Giotto.ai EULA](https://www.giotto.ai/eula) (primary)
8. [VentureBeat on Kimi K3 weights and license](https://venturebeat.com/technology/kimi-k3s-full-weights-are-here-but-theyre-open-with-a-caveat-what-enterprises-should-know)
9. [KI Woche on Giotto and Kimi](https://kiwoche.com/1396/giotto-ai-schweizer-reasoning-modelle-solo-alto-souveraene-ki-kimi/)

Author: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Analysis](https://newruntime.com/analysis.json): Open the latest authorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
