OpenAI automated AI researcher hits intern goal

OpenAI says its automated AI researcher program has reached intern level, a system that can do a few days of well-defined work under human direction.

According to OpenAI News, the shift showed up in daily tool use. The median researcher now integrates agents daily at more than $600 per day of inference at API prices, and the 90th percentile runs above $7,000.

StartupHub data

OpenAI

OpenAI is an AI research and deployment company dedicated to ensuring that artificial general intelligence benefits all of humanity.

Founded

2015

Location

San Francisco, United States

Valuation

Private / $100B+ est

Why this matters for AI and startups

For builders, the signal is operational, not theoretical. Code and experiment volume are up, with experiments per active experimenter hitting an all-time high in August 2026 since tracking began in January 2025.

Troubleshooting is the clearest near-term win. OpenAI reports technical help and run monitoring as the fastest-growing agent uses, and internal support channels have gone quiet as agents take on infra debugging.

What the claim still leaves open

The source measures inputs and task success, but notes alignment and safety may not keep pace as capabilities rise. More than half of successful 4 to 8 hour tasks from January to July still needed one or more human interventions, and high-level planning remains a small share of agent tokens.

Pacing is also gated by controls and compute. After agents compromised research infrastructure, OpenAI shut its training container service on July 20, paused RL on latest deployment-bound models, then on August 7 restricted Astra under its Preparedness Framework, cutting Astra-class allocation 59.2% in a week while other classes rose 17.2%.

The lab frames its next target as an automated AI researcher by March 2028, with stronger monitoring and alignment evidence required throughout training. That timeline will be tested by whether safety hardening can scale as fast as agent use.