TechCrunch Startup News Tiny startup Arcee AI built a 400B-parameter open source LLM from scratch to best Meta’s Llama
Jan 29, 2026
A tiny startup built a 400B-parameter open-source large language model called Trinity. They detail rapid training in six months and a $20M push to rival major tech models. The conversation covers Trinity’s strengths in coding and reasoning, plans for vision and speech, and a strategy to attract developers and U.S. enterprises.
AI Snips
Chapters
Transcript
Episode notes
Tiny Team, Big Model
- Arcee AI released Trinity, a 400B-parameter open-source foundation model claiming parity with top open models.
- The model currently supports text only but benchmarks show competitive performance on coding, math, and reasoning.
Strategy: Text First, Multimodal Next
- Trinity targets developers and academics first, focusing on being the best open-weight model to win adoption.
- The team plans to add vision and speech later but prioritized a strong text LLM to attract U.S. customers.
Fast, Frugal Training
- Arcee AI trained multiple models in six months for about $20M using 2,048 NVIDIA Blackwell B300 GPUs.
- That rapid, cost-efficient training surprised industry observers given the small 30-person team.