Introducing Quasar 1.0

Quasar 1.0 is here. It is a fine-tune of Qwen3.5-0.8B by Alibaba, made by Convergent, and the strongest model we have measured in Convergent Bench 3.0.

Highlights

• 89.2% on Math · Basis (Andromeda-1: 11.1%)
• 84.2% on Knowledge work · Basis (Andromeda-1: 21.8%)
• 75.4% on Agentic coding · Bonus (Andromeda-1: 21.6%)
• 60.0% on Question answering · Basis (Andromeda-1: 34.0%)

What we did

Most of Quasar's skill comes from Qwen3.5-0.8B. Our fine-tune gives it the Quasar identity and a chill, friendly personality in German and English, trained in about four minutes on one RTX 3060 Ti. It has 752 million parameters and was not trained from scratch.

Access

Quasar 1.0 is not public yet. Ask for access in our Discord.