Swiss Lab Giotto Posted Vendor Scores for Its Sovereign Alto and Solo Models. There Is No Model Card, No Independent Board, and No Parameter Count.
Giotto, a Lausanne lab, markets Alto, an EU-hosted consumer app, and Solo, an on-premise single-GPU pitch, with a homepage benchmark table that puts Alto at 100.0% on AIME 2026, 91.9% on GPQA-Diamond, and 52.0% on HLE beside closed rivals. There is no model card, no weight download, and no parameter count on the site, and its terms and legal pages return 404. Secondary aggregators say Alto identified itself as Kimi and that the CEO acknowledged inheriting some Kimi weights; we found no primary for either claim. We read the homepage, which is the only primary there is.
By T.W. Ghost
Giotto, a lab in Lausanne, is marketing two models under a sovereignty banner: Alto, a consumer app hosted in the EU, and Solo, pitched for on-premise and single-GPU deployment. The homepage carries a benchmark table that sets Alto beside closed frontier models and gives it 100.0% on AIME 2026, 91.9% on GPQA-Diamond, and 52.0% on Humanity's Last Exam.
Everything on that table is Giotto's. There is no model card, no weights to download, no parameter count, and no evaluation by Artificial Analysis, an arena, or anyone else. The site's terms and legal pages returned 404 when we tried them, so we cannot say what the license is or whether it restricts publishing benchmarks, which a rumor suggests. A second rumor is the one that will travel: a secondary aggregator says Alto identified itself as Kimi in community tests and that CEO Aldo Podestà acknowledged inheriting some Kimi weights after retraining. We did not find Podestà's own words or a Giotto statement, and we do not print the claim as fact. We read the homepage. Here is what held up, which is the short version.
What Is on the Site, and What Is Not
| Giotto Alto and Solo | |
|---|---|
| Products | Alto, consumer app, EU-hosted; Solo, on-premise and single-GPU |
| Vendor scores, Alto | AIME 2026 100.0%; GPQA-Diamond 91.9%; HLE 52.0% |
| Comparators on the table | Closed frontier models, Giotto's choice |
| Model card | None found |
| Weights, parameter count | None found |
| Terms and license | Pages return 404 |
| Independent score | None |
| Base model | Not stated; Kimi inheritance alleged in secondary reports only |
Sovereignty of hosting is not sovereignty of weights. Data staying in Europe is a verifiable property of a service. The origin of the model serving that data is a separate question, and Giotto's site does not answer it. We have seen the pattern before this year with Multiverse's Quasar and its GLM base: European marketing, external foundation.
A perfect AIME score is a vendor table entry. 100.0% on a 2026 competition set, printed by the lab that built the model, beside rivals the lab picked, with no effort setting and no harness named, is a claim. It may be true. Nobody outside can check it.
The license catch is on hold. A no-benchmarking clause would be a real story. We cannot read the terms, so we do not have one.
Verified vs Unconfirmed: The Scorecard
| Claim | Verdict |
|---|---|
| Alto and Solo marketed on giotto.ai from Lausanne | Verified (homepage) |
| Alto AIME 2026 100.0%; GPQA-Diamond 91.9%; HLE 52.0% | Vendor (homepage table) |
| No model card, weights, or parameter count | Verified absence (site pass) |
| Terms and legal pages 404 | Verified (fetch) |
| Alto self-identifies as Kimi; CEO acknowledged Kimi weights | Unverified (secondary aggregator only) |
| CEO essay "Open Source is a Trojan Horse" | Unverified (not opened) |
| No-benchmarking license clause | Unverified (terms unreadable) |
Who Should Use It
Evaluate Alto or Solo yourself if EU hosting or on-premise deployment is the requirement, and treat the homepage table as a vendor's claim until a board you did not build scores it.
Skip the Kimi story in either direction until Giotto or its CEO says something on the record; it is an allegation about provenance, and provenance is the thing this product is selling.
The frontier is elsewhere: the open models with independent scores this month are on Artificial Analysis's v5 board, and Alto is not one of them. Our model comparison has the current standings side by side.
Sources
*Not sure whether a sovereign-hosted model is sovereign in the way your procurement rules mean? Take the free 2-minute quiz and get matched. Then read about the open model that published its scores and its card, and still only its own: Nanbeige4.3-33B.*