Glossary term
Frontier Model
The most capable AI models at any moment — the handful trained at the largest scale by a few labs, whose release sets the pace for everyone else. A new frontier generation decides how much compute the industry orders next, which is why the term matters to markets, not only to researchers.
AI-generated — produced automatically by Closelook’s systems under this site’s editorial policy.
What it means
“Frontier” describes position, not architecture: the models at the edge of what is possible in reasoning, coding and autonomy, trained on the largest clusters. In 2026 the frontier is set by a small number of labs — Anthropic, OpenAI, Google DeepMind, xAI and Meta among them — each releasing a new generation roughly every year, with smaller, faster models in between. Regulators use the word to define which models face the strictest evaluation before release.
Each frontier generation costs several times more compute to train than the last and, once released, more compute to run at scale. The cadence of frontier releases is therefore the demand curve of the AI build-out.
Why it matters for the AI trade
Orders for chips, memory, optical components and power are placed against the next frontier generation. When the three largest labs said on 12 September 2026 that the frontier should move more slowly and be evaluated by outsiders — the pacing call — the market repriced the suppliers in a session, while the companies that buy frontier models to build products rose. Whether the frontier keeps its cadence is the fact every AI supplier’s valuation now depends on.
How Closelook uses it
The recursive self-improvement read tracks what the frontier labs have actually shipped against the rumours; the economic scenarios entry covers what a faster or slower frontier does to wages and capital income. Frontier release dates sit in the Morning 10 calendar next to earnings.
Common questions
- Which models are frontier models in 2026?
- The latest large releases from Anthropic, OpenAI, Google DeepMind, xAI and Meta, plus a small number of others at comparable scale. The set changes with each release cycle; the label follows capability, not brand.
- Why does a frontier release move chip stocks?
- Because training the next generation and serving it at scale requires the next round of orders. A release on schedule confirms the demand curve; a delay or a deliberate slowdown reprices it.
- Is a frontier model the same as a foundation model?
- A foundation model is any large general-purpose model that others build on; a frontier model is one at the current limit of capability. All frontier models are foundation models; most foundation models are not at the frontier.