Colosseum Models by iGenius

iGeniusLlama 3 CommunityOpen weights
2 models2025Up to 16K ctx

Last refreshed 2026-10-09. Next refresh: weekly.

Details

ResearcheriGenius
LicenseLlama 3 Community
Commercial useRestricted
Models2
Released2025
Max context16K

About

Colosseum is a large language model family from Domyn, the Italian AI company formerly called iGenius, built for enterprises in highly regulated sectors such as financial services and public administration. Its flagship, Colosseum 355B, was introduced in January 2025. It came from continued pretraining on NVIDIA DGX Cloud: Domyn added layers to a foundation model to reach 355 billion parameters, extended the context length from 4K to 16K tokens, and ran part of the training in FP8. Its training data spans more than 50 languages with an emphasis on European languages such as Italian, plus finance and reasoning sources, and the model was then aligned with supervised fine-tuning and Direct Preference Optimization to improve instruction following.

Decision facts

Best fit
coding
Capability starting point
Colosseum 355B with 16K context
Lowest tracked input
Not tracked
Closest related family
Italia

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

2 in view
Colosseum 355BCurrent

Use when the workload needs 16K context and 355B parameters.

2025-0316K context355B parameters
Colosseum 355B InstructCurrent

Use when the workload needs 16K context and 355B parameters.

2025-0116K context355B parameters

Release Timeline

2 release groups
2025-03
1 current
Colosseum 355B
16K context355B parameters
Current
2025-01
1 current
Colosseum 355B Instruct
16K context355B parameters
Current

Specifications(2 models)

Colosseum model specifications comparison
ModelReleasedContextParameters
Colosseum 355B2025-0316K355B
Colosseum 355B Instruct2025-0116K355B

Available From(1 provider)

NVIDIA NIM

Popular comparisons in this family