OpenAI's "Automated Research Intern": 3.1 Agent Workdays per Human — The Cost Calculation Behind the $7,000 Day
TL;DR: OpenAI reports a productivity milestone for its coding and research agents: By mid-August 2026, the agents are delivering 3.1 agent workdays per human workday. With a median daily spend of $600 and peak values exceeding $7,000, this results in a billable target cost range of $200 to $2,260 per generated workday — with an officially set final sprint leading up to March 2028.
What the Automated Research Intern Is
The Automated Research Intern is OpenAI's term for its autonomous research and coding regime that accelerates the daily workflow of developers. The core metric is the Agent Workday Index (AWD): It compares the workdays completed by agents to the human workdays affected by them.
An index of 3.1 specifically means: For every paid human workday utilized, agents produce the equivalent of 3.1 workdays of output. This shifts the bottleneck from typing to thinking — humans define problems, and agents process them.
The September 6, 2026 Milestone
OpenAI documented the current status of 3.1 AWD in its Research Acceleration Report on September 6, 2026. The index has been rising almost linearly since the introduction of the agent product line; the last three observation months were at 2.4, 2.7, and now 3.1.
The official goal: By March 2028, the index is projected to reach 6.0, meaning six agent workdays per human day. Secondary sources (unite.ai, cellcog, DataStudios) congruently confirm this sequence of numbers and the timeframe.
The Cost Calculation Behind the $7,000 Day
The AWD figure is only half the story — the crucial factor is the dollar value per generated workday. Assuming a median daily spend of $600 per human and top 10 percent values exceeding $7,000, the following calculation emerges:
| Model | Daily Spend (USD) | AWD Index | USD per Agent Workday |
|---|---|---|---|
| Median | 600 | 3.1 | ≈ 194 |
| Top 10 Percent | 7,000+ | 3.1 | ≈ 2,258 |
This calculation shows: A team operating at the median level receives outputs for less than $200 for every agent workday. Teams in the top 10 percent amortize their agent costs after just a few days of intensive use.
Why This Is Relevant for Teams
The productivity leap is no longer a parallel phenomenon: In OpenAI's own reports, the AWD index has been growing in sync with the number of active agent sessions over the past quarter. This means that acceleration scales with usage, rather than with team growth.
| Observation | Practical Implication |
|---|---|
| AWD 3.1 per human | One human manages more than three daily workloads |
| Median 600 USD/day | Cost level for most teams |
| Top 10 % > 7,000 USD/day | Power users reach 12x the daily value |
| Target March 2028: 6.0 AWD | Doubling planned in about 18 months |
FAQ
How should the Agent Workday Index be interpreted? The index compares the total agent-delivered workdays with human workdays in the same period. A value of 3.1 means: 3.1 agent days of output are generated per human day. The index is therefore a multiplier, not an absolute number of hours.
What is the actual cost of an agent workday? Measured at the median, around $194, calculated from a $600 daily spend divided by 3.1. For the top 10 percent of users, the value is around $2,258 per agent workday. Both figures refer to the documented OpenAI report series from September 2026.
Which sources confirm the numbers? The primary source is the OpenAI Research Acceleration Report from September 6, 2026. Secondary confirmations with an identical index trend come from unite.ai, cellcog.ai, and datastudios.org. There are no discrepancies regarding the exact index figures between the sources.
What are the limits of the index? The AWD index counts delivered output days, not their quality per issue. An agent day can contain many small or a few large tasks. For budget planning, the index should therefore be combined with your own ticket throughput metrics.
What will change by March 2028? OpenAI is targeting an index of 6.0 for March 2028, representing a doubling compared to 3.1. If this occurs linearly, the median price per agent workday at the same daily spend would be close to $100. Teams should adopt standard weekly audits of their agent usage until then.
Sources
- https://openai.com/index/research-acceleration-view-inside-openai
- https://www.unite.ai/openai-hits-goal-of-building-an-automated-research-intern
- https://cellcog.ai/blog/openai-automated-research-intern
- https://www.datastudios.org/post/openai-automated-research-intern-coding-agents-research-acceleration-ai-researcher