Giotto AI announced the Giotto family of models, unveiling Giotto Alto and Giotto Solo, "the smartest European sovereign AI models on the market." The company calls itself "a Swiss lab with a European heart" that built them "so that Europe can run advanced AI on its own terms." Both are usable at cloud.giotto.ai.
The two models
- Giotto Alto: the flagship, built for advanced reasoning, free for everyone and available on the App Store and Google Play. It is hosted on European cloud infrastructure, so users' data stays in Europe. Coverage reports it runs on eight B300 GPUs.
- Giotto Solo: the smartest model that runs on a single GPU, so organizations can keep their AI on their own infrastructure. Launch coverage lists deployment options across shared cloud hosted and operated in Europe, dedicated cloud in isolated environments at partner data centers in Switzerland or elsewhere in Europe, and on-premises installation through partner integrators. Coverage of the license describes it as limiting use to noncommercial testing and evaluation and barring reverse engineering and publishing benchmarks without written consent, so these are not open-weight models.
Giotto's benchmark charts
Giotto's benchmark tables, from the launch thread.
Alto is compared with frontier models. It scores 100.0% on AIME 2026, level with GPT 5.5 (xhigh) and Claude Opus 4.8 (max) and above Gemini 3.1 Pro preview (98.3%), Kimi K3 (96.7%) and Aleph Alpha's Kolibri-1 (96.0%), and 99.80% on Math-500 against 99.40% for GPT 5 (high) and 99.10% for Claude Sonnet 4. It does not lead the harder science benchmarks: on GPQA-Diamond, Alto's 91.9% trails GPT 6 Astra (96.1%), Gemini 3.8 Flash (95.3%), Claude Fable 5.1 (93.7%) and Kimi K3 (93.5%), and on Humanity's Last Exam its 52.0% is behind Claude Fable 5.1 (59.1%) and GPT 6 Astra (54.7%), though ahead of Gemini 3.8 Flash (47.8%), Kimi K3 (46.9%), Mistral Large 4 (35.0%) and Kolibri-1 (21.5%).
Solo is compared with open models of its size class and leads every row shown: 100.0% on AIME 2025 (Nemotron 3 Nano 30B-A3B 99.2%, GPT-OSS-120B 97.9%, Ministral 3 14B 85.0%), 96.6% on AIME 2026 against 89.2% for Gemma 4 31B, 88.3% on GPQA-Diamond against 84.3% for Gemma 4 31B and 80.1% for GPT-OSS-120B, 37.5% on HLE against 26.5% for Gemma 4 31B and 19.0% for GPT-OSS-120B, and 99.80% on Math-500 against 94.30% for R1-Distill-Qwen-32B.
A Community Note attached to the post says the graphic shows self-reported scores compared with specific variants of other models, and that independent leaderboards as of late October 2026 list no Giotto and show different leading scores on AIME 2026, GPQA-Diamond and HLE. Several of the comparison entries are older models, such as GPT 5 and Claude Sonnet 4 on Math-500.
The manifesto
Giotto also published a manifesto, "why we build sovereign AI." Its argument is that as AI becomes infrastructure, "the conditions under which we can use it will matter as much as its capabilities," and that people can gain access to extraordinary intelligence while losing the ability to decide where it runs, what information they entrust to it and whether they can take it elsewhere. "We refuse to accept that bargain."
Giotto says Europe must build an alternative that can compete globally, and that "a European address disguised as sovereignty is not enough": sovereignty is "a freedom to exercise," meaning genuine control over the technology people depend on. It describes Giotto as "an AI model and operating system to make advanced reasoning efficient to run on infrastructure you choose," needing less compute, energy and infrastructure, and sums up its logic as "efficiency makes portability possible; portability makes sovereignty real." The text closes on a commitment to build "for generations to come" with attention to the impact on people, communities and shared resources, and five principles: powerful where it matters, efficient by design, private on your terms, sovereign by nature, and intelligence you control.
Context
Giotto is based in Lausanne, according to launch coverage, and joins a European push for sovereign AI, with Aleph Alpha's Kolibri-1, Mistral's Large 4 and OVHcloud's planned frontier models among the others.