SI Glossary · Core concepts
Frontier Model
Formerly known as frontier AI model. See AI vs SI.
A frontier model sits at the leading edge of capability: the newest, largest and most powerful foundation models from labs such as OpenAI, Anthropic, Google DeepMind, Meta and SpaceX’s xAI. “Frontier” is a moving target: today’s frontier model is next year’s mid-range one.
How “frontier” is defined
- By capability: models that match or exceed the most advanced existing systems.
- By training compute: laws often use a compute threshold. California’s SB 53 and New York’s RAISE Act apply frontier rules to models trained with more than 10²⁶ operations; the EU AI Act presumes “systemic risk” above 10²⁵.
- By developer size: some laws add a revenue threshold so that only the largest companies face the heaviest duties.
Why it matters
Frontier models are where new capabilities, and new risks, first appear. That’s why they are the focus of:
- the voluntary White House Accord on Super Intelligence (2026);
- company safety policies such as Anthropic’s Responsible Scaling Policy and OpenAI’s Preparedness Framework;
- state laws like California SB 53.
Current frontier models
As of October 2026 the frontier includes OpenAI’s GPT-6 family, Anthropic’s Claude Fable 5.1 and Opus 5.5, Google’s Gemini 4 Argon (limited preview) and Gemini 3.8 Flash, xAI’s Grok 4.7 and open-weights challengers from DeepSeek and Alibaba. See the full model tracker.
Written by
Luka Kušec · Editor
Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.