The AI lab says its models went from 0% to 26% of next-generation development work in 2026, and it expects that share to keep climbing

Anthropic projects that its Claude models will lead 80-90% of frontier model training in 2027.

According to the company, Claude’s contribution to next-generation model development stood at 0% at the beginning of 2026. By April, that figure had reached 3%.

Then it jumped. Anthropic indicated that Claude’s share hit 26% by August 2026.

The 80-90% target for 2027 is a projection, not a done deal.

Claude went from producing negligible code to writing 50% of Anthropic’s internal code shortly after a major model release in August 2025. The company then aimed for 80-90% by the end of that year.

On October 2, 2026, the company announced the Claude Frontier Academy, a $100 million initiative. Its goal is to train 10,000 “Frontier Deployed Engineers” by the end of 2027.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

The program uses a residency-style curriculum, similar to how doctors learn on the job under supervision. The first certifications are expected by early 2027.

Anthropic is working with major firms including Accenture and Morgan Stanley on the effort. Those partnerships are meant to give trainees real-world placements rather than just classroom hours.

Anthropic is aiming for approximately 9.5 gigawatts of compute capacity by the end of 2027. To get there, it is leaning on partnerships with cloud providers including AWS and Google , among others.

Anthropic was founded by former OpenAI researchers and brands itself as an AI safety-focused laboratory.

It also means the industry will be watching how Anthropic measures and reports these figures. Percentages like 26% depend heavily on how “contribution” is defined, and outsiders cannot easily verify internal metrics.

For enterprises, the Frontier Academy may be the most practical piece of news. A pipeline of 10,000 certified engineers could make it easier for large firms to deploy Claude at scale, assuming the program hits its targets and the first certifications land in early 2027 as expected.

The key things to watch are whether Anthropic publishes updated contribution figures in the coming months, how it defines what counts as Claude “leading” training, and whether the compute buildout stays on schedule.