Yield in AI usage.
Yield measures the productive output share of total token flow in an AI operator session. It answers a simple question: of all the tokens that moved through the system, what fraction became output? It is one of five canonical derived metrics in the MO§ES™ evaluation framework.
The formula.
Yield is the ratio of output tokens to the total token flow — input, output, reused context, and newly built context combined. It captures what fraction of the entire token budget became productive generation.
Where:
- O = Output (tokens generated by the model)
- I = Input (fresh tokens sent to the model)
- R = Cache Read (reused context tokens)
- W = Cache Write (new context built tokens)
Yield ranges from 0 to 1. A yield of 0.25 means that one quarter of all tokens that moved through the session were output. A yield of 0.10 means only one tenth were output — the rest was input, reused context, or newly built context. The metric normalizes across sessions of different sizes, making it comparable across operators who handle different volumes.
What yield tells you.
Yield is the efficiency lens of the canonical metric set. It describes how much of the total token investment converts to generated output, regardless of whether that output is correct, useful, or valuable.
A larger share of total token flow is output. The operator's sessions produce more generation relative to the total token budget consumed. This may reflect efficient prompting, concise context management, or workflows that prioritize generation over context setup.
A smaller share of total token flow is output. More tokens are spent on input, context reuse, or context construction relative to generation. This may reflect heavy context-building workflows, exploratory sessions, or multi-turn debugging patterns.
Yield does not measure output quality. An operator can have high yield with low-quality output, or low yield with excellent output that required significant context investment. The metric is most useful when combined with leverage and construction to understand the operator's overall context economy.
Yield in the evaluation pipeline.
MO§ES™ computes yield from canonical telemetry for every observed session. The metric enters the pipeline at three stages.
Computed per session from all four canonical telemetry signals. No prompt content required — only token counts.
Operators are ranked by yield within their cohort. Percentile bands show where each operator falls relative to peers under the same workflow and model conditions.
Yield contributes 30% of the AI Operator Development Index, the 0–100 composite score that summarizes overall operator behavior. It is weighted equally with leverage.
In the synthetic demo cohort of 50 operators across 1,668 observations spanning 5 AI providers — ChatGPT, Claude, Codex, Copilot, and Cursor — yield values cluster around a median with a meaningful right tail. The top 10% of operators show yield values substantially above the cohort median, suggesting more efficient token-to-output conversion. These are structural observations from synthetic data, not validated performance findings.
What yield does not tell you.
Yield is a derived metric labeled DEVELOPMENTAL. It is a structural efficiency signal whose relationship to actual performance outcomes is still being tested.
- Not a quality measure. High yield means more of the token budget became output. It does not mean the output was correct, useful, or valuable.
- Not a productivity score. Yield measures token efficiency, not work output. An operator with low yield may produce higher-quality work that requires more context investment.
- Not an employee ranking. DEVELOPMENTAL labels mean results route workflows, not people. No adverse employment actions are permitted from pilot data.
- Workflow-dependent. Different task types have different natural yield ranges. Code generation, exploratory analysis, and debugging have inherently different token economies.
- Validation required. The relationship between yield and validated business outcomes has not been independently established. Correlations are labeled ASSOCIATION, never CAUSATION.
Every yield measurement carries provenance: source telemetry window, operator cohort, evidence label (DERIVED), decision-use label (DEVELOPMENTAL), and synthetic-data flag where applicable.
Read alongside.
How much context the operator reuses and builds relative to new input. (R + W) / I.
Output relative to input and reused context. O / (I + O + R).
The 0–100 AI Operator Development Index. Yield contributes 30%.