Agent Cost & UsageAgent Cost Display Precision: Why a Line Item Can Show $0.00

Summary

A $0.00 value in Agentic → Cost → Agent Cost Items does not necessarily mean that cost calculation or SDK instrumentation failed.

There are two common expected reasons a line item can show $0.00:

1. The calculated line-item cost is smaller than the precision displayed in the Mavvrik UI.

2. A provider reports a usage breakdown for observability that is already included in another billable token category. For example, OpenAI reasoning tokens are reported as a breakdown of output tokens and should not be charged again as a separate line item.

These are expected behaviors and should not be treated as bugs by themselves.

Where this applies

This behavior can be seen in the Cost view for agentic telemetry captured through the Mavvrik SDK or other supported agent observability integrations.

Typical examples include:

• embedding operations with a small number of tokens;

• inexpensive models with very small token counts;

• individual line items before enough usage has accumulated to produce a cost visible at the UI precision;

• reasoning-token breakdowns reported by supported OpenAI reasoning models.

Cost display precision

Mavvrik displays very small non-zero costs to a maximum of five digits after the decimal point in the Agent Cost Items table.

The smallest cost represented at that precision is:

$0.00001

Mavvrik does not extend the Cost column to a sixth decimal place. Therefore, if a calculated line-item cost is below the five-decimal display precision—for example:

$0.000001

—the table displays:

$0.00

The displayed $0.00 means that the individual cost is too small to be represented at the current UI precision. It does not, by itself, mean that the usage event was missed or that the SDK failed to capture the operation.

Important: Treat this as a display-precision behavior. Do not infer a telemetry or pricing defect from a single $0.00 line item when the usage quantity and model information are present.

Example: low-cost embedding usage

For the same embedding model, different token quantities can produce different displayed costs:

This explains why line items using the same provider and model can show $0.00, $0.00001, or $0.00002 depending on the quantity consumed.

OpenAI reasoning tokens: why the reasoning line can show $0.00

A separate but related case can occur when an agent uses an OpenAI reasoning model such as:

• Provider: openai

• Model: gpt-5.4-mini-2026-03-17

OpenAI exposes reasoning-token usage separately so that applications and observability platforms can inspect how much of the generated output was used for reasoning. In the OpenAI API usage object, reasoning_tokens is reported inside the detailed breakdown of output_tokens.

In other words, reasoning tokens are a subset of the generated output/completion tokens, not an additional token category that should be added on top of output tokens for billing.

Because of this, Mavvrik may show the reasoning-token quantity as its own observable usage line while displaying its standalone cost as:

$0.00

This is expected. The output-side model cost is attributed to the completion/output tokens. Mavvrik does not apply the output-token price a second time to the reasoning-token breakdown, because doing so would double-count the same usage.

Important: A non-zero reasoning_tokens quantity with a $0.00 reasoning-line cost does not mean reasoning usage was ignored. The reasoning usage is visible for observability, while its cost is already represented through the parent completion/output-token usage.

Example

Assume an OpenAI gpt-5.4-mini-2026-03-17 request reports:

• prompt_tokens: 500

• completion/output_tokens: 1,200

• reasoning_tokens: 800

The 800 reasoning tokens are included within the 1,200 output tokens. They should not be interpreted as 1,200 + 800 = 2,000 independently billable output tokens.

A Mavvrik view can therefore look conceptually like this:

So if the customer sees a non-zero reasoning token quantity but $0.00 cost for that reasoning row, while the completion/output-token row carries the output cost, the behavior is correct.

OpenAI reference behavior

OpenAI's API documentation defines output_tokens_details.reasoning_tokens as a detailed breakdown of the response's output_tokens. OpenAI also prices gpt-5.4-mini using input and output token rates rather than a separate reasoning-token price.

References:

• OpenAI API — GPT-5.4 mini model

• OpenAI API — Responses usage and output token details

How to verify that the line item is valid

When a customer reports a $0.00 agent cost line item, first verify the surrounding telemetry:

1. Confirm that Quantity is greater than zero.

2. Confirm that the expected Usage Metric is present, such as prompt_tokens, completion_tokens, or reasoning_tokens.

3. Confirm that Model Provider and Model Name are populated as expected.

4. Confirm that the expected Operation is recorded.

5. Identify whether the row is a normal billable usage category or a provider-reported usage breakdown such as reasoning_tokens.

6. Compare the row with other line items for the same model and request.

For ordinary token or embedding usage, if the surrounding fields are populated and only the displayed cost is $0.00, the most likely explanation is that the amount is below the UI display precision.

For OpenAI reasoning_tokens, a $0.00 standalone cost can be expected even with a large quantity because the reasoning tokens are already included in output/completion-token usage and are not priced again separately.

When to investigate as a potential defect

A $0.00 display alone is not sufficient evidence of a defect. Investigate further when one or more of the following is also true:

• token or usage quantity is missing when usage is expected;

• model or provider attribution is missing or incorrect;

• the operation is missing or incorrectly classified;

• ordinary billable usage with larger quantities that should produce a visible cost continues to show $0.00;

• an OpenAI reasoning request reports completion/output usage but the expected output-token cost is also missing;

• reasoning-token quantity appears to be added on top of completion/output tokens for cost calculation, resulting in double-counting;

• related totals or aggregated costs are inconsistent beyond what can be explained by display precision or provider usage-breakdown semantics;

• the expected model pricing configuration is missing or incorrect.

Customer-facing explanations

Very small calculated cost

Mavvrik is still capturing the usage for this operation. The individual calculated cost is extremely small and falls below the precision displayed in the Cost column. The UI displays costs up to five decimal places, so a value such as $0.000001 is shown as $0.00 rather than extending the table to a sixth decimal place. As token or usage quantity increases, the cost becomes large enough to appear as values such as $0.00001 or $0.00002.

OpenAI reasoning tokens

OpenAI reports reasoning tokens separately as a breakdown of generated output usage. Mavvrik keeps that reasoning-token quantity visible for observability, but it is already included in the completion/output-token usage used for cost calculation. For that reason, the reasoning-token row can correctly show $0.00 while the corresponding completion/output-token row carries the model's output cost. Charging the reasoning row separately would count the same output usage twice.

Related documentation

• Agent Cost & Usage

• Cost & Attribution

• Troubleshooting Agent Data