Key Takeaways
- Claude now leads 26% of Anthropic’s AI research and development work.
- The figure rose from zero in February.
- About 90% of Anthropic’s research now involves Claude.
Anthropic said its Claude AI model now leads 26% of the company’s Claude AI research and development work, marking a sharp rise in AI-led model development.
Claude Takes a Larger Role in AI Development
Anthropic said Claude led 26% of its model research and development work in August. In these tasks, Claude can complete most of the work from a high-level instruction, while people provide direction and supervision.
The figure was zero in February, showing how quickly the model’s role has expanded. Anthropic said the measure comes from an internal review of its research tasks and aims to track how much work AI systems can handle in the development of future models.
Claude does not yet work fully on its own. Human researchers remain involved in setting goals, reviewing results, and overseeing the system’s actions. Anthropic said this distinction is important because the company does not consider its current work to be fully autonomous AI development.
The company also said about 90% of its research and development work now involves Claude. In these cases, Claude handles large parts of tasks while people remain closely involved in the process.
Anthropic Tracks AI-Led Research
The company is using these figures to measure how quickly Claude AI research systems are taking on more complex development tasks. Anthropic said that tracking the numbers can help researchers understand how close AI systems may be to carrying out more of their own development.
The company has also reported wider gains from Claude in its engineering work. Anthropic said that more than 80% of the code it merged into its codebase was authored by Claude as of May 2026. It also said the typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024, while noting that lines of code are an imperfect measure of productivity.
Claude has also become more capable at running defined experiments. Anthropic said its internal model achieved about a 52-fold speedup when optimizing code for a model-training task in April, compared with about three times for an earlier model in May 2025.
These gains are changing how Anthropic develops and tests its own models. The company now uses Claude to review code for bugs, security problems, and other defects before engineers merge changes into its codebase.
Human Oversight Remains Central
Anthropic said around 30,000 AI agents were active on its internal research and engineering platform in August. The company uses monitoring systems to review their actions and block activity that does not meet its safety requirements.
In August, more than 90% of Anthropic’s research involved human-AI collaboration, showing that people still play a major role across the company’s development work. The company said every agent action goes through screening, while its internal data also tracks how often safeguards block actions. Human researchers remain central to Claude AI research, setting goals, reviewing results, and monitoring the actions of AI agents.
Anthropic has called for AI companies to share more information about how much of their research AI systems can perform. The company said common measures could make it easier to track progress across AI labs.
The latest figures come as major AI companies focus on systems that can handle longer and more complex tasks. Anthropic’s data provides a measure of how quickly Claude is moving from assisting researchers to taking the lead on parts of AI model development.
