I think there must be a bug in your Usage portal related to file search storage. In the second usage chart below, it shows that we are using 133GB each day. In the first taken from the Spend categories, you’ll see that you’re charging about $0.52 for the same day. In reality, I can believe that we have about 5.2GB of storage – which would be consistent with the $0.52 per day. My guess is that there is just a bug in the usage chart.
Let’s take a look at your maximum day:
>>> usage_max=153_383_192_439
>>> print(usage_billed_gb:=(usage_max/2**30)-1)
141.84922968503088
>>> print(cost:=usage_billed_gb*0.10)
14.184922968503088
Let openai know to increase your daily bill to $14?
I can pile on. I’ve got vector stores in megabyte-hours:
Go to the depicted link:
https://platform.openai.com/usage/vector-stores
We can look another way: Use the organization admin API.
I shouldn’t get any “costs” line items for vector stores on the costs endpoint, as I’m well under 1GB.
However, If the vector store usage endpoint was supposed to return “current bytes” and not “billed bytes”, even bytes up-to-the-minute by parameter … that’s not getting done for me either.
API admin endpoint code for you to try, that will probably say the same “bytes” as the usage page. If so, you might need to ultimately iterate through the size of all vector stores in a vector store list - if the listing’s not broke also.
Report if you get nothing or problems with the printing in this code, as I have nothing to print. (Try “group_by”: [“project_id”] in get_recent_vector_usage()to break the printing too).
'''OpenAI admin framework – vector store usage and API costs printer
Environment:
export OPENAI_ADMIN_KEY=... # required “sk‑admin‑…” key
Dependencies:
pip install httpx (there if you have openai module)
'''
import json
import logging
import os
from datetime import UTC, datetime, timedelta
from typing import Any
import httpx
logger = logging.getLogger(__name__)
def admin_get(
path: str,
*,
params: dict[str, Any] | None = None,
timeout: float = 120.0,
) -> tuple[httpx.Response, dict[str, str]]:
"""GET wrapper for query-only OpenAI endpoints (no body/JSON)"""
api_key = os.getenv("OPENAI_ADMIN_KEY")
if not api_key:
raise ValueError("Set OPENAI_ADMIN_KEY first (must be an *admin* key).")
try:
resp = httpx.get(
f"https://api.openai.com/v1/{path}",
headers={"Authorization": f"Bearer {api_key}"},
params=params, timeout=timeout,
)
hdrs = dict(resp.headers)
resp.raise_for_status()
return resp, hdrs
except httpx.HTTPStatusError as err:
body_text = err.response.text.strip() or "<no body>"
print(f"[Admin API] HTTP {err.response.status_code}: {body_text}", flush=True)
logger.error("HTTP %s: %s", err.response.status_code, body_text)
raise
except httpx.RequestError as err:
logger.error("Request error: %s", err)
raise
def get_recent_costs(
past_days: int = 14,
) -> dict[str, Any]:
"""
Retrieve organisation‑level cost daily buckets up to today.
Parameters
----------
past_days
Number of 24‑hour buckets (1 – 180).
Returns
-------
Costs object with result buckets, amended with datetime. Example:
{
"object": "page",
"data": [
{
"object": "bucket",
"start_time": 1763337600,
"end_time": 1763424000,
"results": [
{
"object": "organization.costs.result",
"amount": {
"value": 0.06, # can also be exponential, ie 6e-05
"currency": "usd"
},
"line_item": null, # When group_by+=line_item
"project_id": null, # When group_by+=project_id
"organization_id": "org-1234"
}
]
}
],
"start_datetime": "2025-11-17T00:00:00Z" # added by function
"has_more": false,
"next_page": null
}
"""
past_days = max(1, min(past_days, 180))
now = datetime.now(tz=UTC)
start_dt = (now - timedelta(days=past_days)).replace(
hour=0, minute=0, second=0, microsecond=0
)
params: dict[str, Any] = {}
params.update(
{
"start_time": int(start_dt.timestamp()), # send UNIX epoch time
"end_time": int(now.timestamp()), # also UNIX epoch time
"limit": past_days, # items before cursor pagination
"bucket_width": "1d", # only "1d"
"group_by": ["line_item"], # ["line_item" and/or "project_id"]
}
)
resp, _ = admin_get("organization/costs", params=params)
costs: dict[str, Any] = resp.json()
# add human-readable start date to the response object
for bucket in costs["data"]:
start_dt = datetime.fromtimestamp(bucket["start_time"], tz=UTC)
bucket["start_datetime"] = start_dt.isoformat().replace("+00:00", "Z")
return costs
def printable_costs(costs: dict[str, Any]) -> str:
"""
Format organisation cost buckets into a Markdown-compatible table.
Parameters
----------
costs
The dict returned by ``get_recent_costs()``.
Returns
-------
str
A Markdown table showing daily costs per line item, auto-width.
Example printable return produced
-------
| UTC Date | Line Item | Amount (USD) |
| -------------------- | ---------------------------------- | ------------ |
| 2025-11-19T00:00:00Z | ft-gpt-4.1-nano-2025-04-14, input | 0.0000 |
| | ft-gpt-4.1-nano-2025-04-14, output | 0.0001 |
| | ft-gpt-4o-2024-08-06, input | 0.0001 |
| | ft-gpt-4o-2024-08-06, output | 0.0002 |
"""
headers = ("UTC Date", "Line Item", "Amount (USD)")
# Track rows as tuples of (date, line_item, amount_str)
rows: list[tuple[str, str, str]] = []
# Initialise column widths from header labels
widths = [len(h) for h in headers]
for bucket in costs.get("data", []):
results = bucket.get("results") or []
if not results:
continue
# Full UTC datetime string (already added in get_recent_costs)
date_str = bucket.get("start_datetime", "")
first_for_day = True
for result in results:
line_item = result.get("line_item", "")
amount_value = result.get("amount", {}).get("value", 0.0)
# Format amount to 4 decimal places
amount_str = f"{float(amount_value):.4f}"
# Only show the date once per day block
date_cell = date_str if first_for_day else ""
first_for_day = False
row = (date_cell, line_item, amount_str)
rows.append(row)
# Update column widths based on this row
widths[0] = max(widths[0], len(row[0]))
widths[1] = max(widths[1], len(row[1]))
widths[2] = max(widths[2], len(row[2]))
# Build Markdown table lines
lines: list[str] = []
# Header row
header_row = "| " + " | ".join(
headers[i].ljust(widths[i]) for i in range(len(headers))
) + " |"
lines.append(header_row)
# Separator row, width-aligned with header/body
separator_row = "| " + " | ".join(
"-" * widths[i] for i in range(len(headers))
) + " |"
lines.append(separator_row)
# Data rows
for date_cell, line_item, amount_str in rows:
line = "| " + " | ".join(
[
date_cell.ljust(widths[0]),
line_item.ljust(widths[1]),
amount_str.rjust(widths[2]),
]
) + " |"
lines.append(line)
return "\n".join(lines)
def get_recent_vector_usage(
past_days: int = 7,
) -> dict[str, Any]:
"""
Retrieve organisation vector store daily bytes buckets up to today (now).
Work in progress.
Parameters
----------
past_days
Number of one-day buckets (1 – 31), default 7 like API.
Returns
-------
object from API
Example Return (if empty results buckets and limit=2 in "1d")
-------
{
"object": "page",
"has_more": true,
"next_page": "page_AAAAAGkf3HXKKd36AAAAAGkeWgA=",
"data": [
{
"object": "bucket",
"start_time": 1763424000,
"end_time": 1763510400,
"start_time_iso": "2025-11-18T00:00:00+00:00",
"end_time_iso": "2025-11-19T00:00:00+00:00",
"results": [],
},
{
"object": "bucket",
"start_time": 1763510400,
"end_time": 1763596800,
"start_time_iso": "2025-11-19T00:00:00+00:00",
"end_time_iso": "2025-11-20T00:00:00+00:00",
"results": [],
}
]
}
"results" contains bucket format items:
{
"object": "organization.usage.vector_stores.result",
"usage_bytes": 1024,
"project_id": "proj_abc" # When group_by=project_id, otherwise null
}
"""
past_days = int(max(1, min(past_days, 31))) # constrain input param, not raise
now = datetime.now(tz=UTC)
start_dt = (now - timedelta(days=past_days)).replace(
hour=0, minute=0, second=0, microsecond=0
)
params: dict[str, Any] = {}
params.update(
{
"start_time": int(start_dt.timestamp()),
"end_time": int(now.timestamp()), # also UNIX epoch time
"limit": past_days, # buckets before cursor pagination
"bucket_width": "1d", # 1m, 1h and 1d ... 1d is max 31 days page limit
"group_by": [], # optional ["project_id"], not demonstrated
# "project_ids": [], # empty array might return nothing?
}
)
resp, _ = admin_get("organization/usage/vector_stores", params=params)
usage: dict[str, Any] = resp.json()
# Un-needed; this API now includes human-readable "start_time_iso"
# Future: We could still proprietary "Zulu" the time to remove the +00:00 timezone
# for bucket in usage["data"]:
# start_dt = datetime.fromtimestamp(bucket["start_time"], tz=UTC)
# bucket["start_datetime"] = start_dt.isoformat().replace("+00:00", "Z")
return usage
def printable_vector_usage(usage: dict[str, Any]) -> str:
"""
Format organisation cost buckets into a Markdown-compatible table.
Parameters
----------
costs
The amended dict returned by ``get_recent_vector_usage()``.
Returns
-------
str
A Markdown table showing daily costs per line item. Human-interpretable.
| UTC Date | Project ID | Usage (bytes) |
| -------- | ---------- | ------------- |
| ...
"""
headers = ("UTC Date", "Project ID", "Usage (bytes)")
# Track rows as tuples of (date, project_id, usage_str)
rows: list[tuple[str, str, str]] = []
# Initialise column widths from header labels
widths = [len(h) for h in headers]
for bucket in usage.get("data", []):
results = bucket.get("results") or []
if not results:
continue
# API returns e.g. "2025-11-19T00:00:00+00:00"
# Make it a short, explicit UTC form by replacing the timezone with "Z"
start_time_iso = bucket.get("start_time_iso", "") or ""
if start_time_iso.endswith("+00:00"):
date_str = start_time_iso.replace("+00:00", "Z")
else:
date_str = start_time_iso
first_for_day = True
for result in results:
# When group_by=["project_id"] this is a real ID; otherwise null
project_id = result.get("project_id")
project_str = project_id if project_id is not None else ""
usage_bytes = result.get("usage_bytes", 0)
try:
usage_int = int(usage_bytes)
except (TypeError, ValueError):
usage_int = 0
usage_str = str(usage_int)
# Only show the date once per day block
date_cell = date_str if first_for_day else ""
first_for_day = False
row = (date_cell, project_str, usage_str)
rows.append(row)
# Update column widths based on this row
widths[0] = max(widths[0], len(row[0]))
widths[1] = max(widths[1], len(row[1]))
widths[2] = max(widths[2], len(row[2]))
# Build Markdown table lines
lines: list[str] = []
# Header row
header_row = "| " + " | ".join(
headers[i].ljust(widths[i]) for i in range(len(headers))
) + " |"
lines.append(header_row)
# Separator row, width-aligned with header/body
separator_row = "| " + " | ".join(
"-" * widths[i] for i in range(len(headers))
) + " |"
lines.append(separator_row)
# Data rows
for date_cell, project_str, usage_str in rows:
line = "| " + " | ".join(
[
date_cell.ljust(widths[0]),
project_str.ljust(widths[1]),
usage_str.rjust(widths[2]),
]
) + " |"
lines.append(line)
return "\n".join(lines)
apicosts = {}
apiusage = {}
def main() -> None:
"""Get and print some OpenAI admin endpoint data,
leaving API responses in globals at at REPL console"""
global apicosts, apiusage
# Get OpenAI daily costs billed to an organization
apicosts = get_recent_costs(past_days=14)
#print(json.dumps(apicosts, indent=2)) # for debugging
print(printable_costs(apicosts))
# New - get the usage from vector store storage billed by day
apiusage = get_recent_vector_usage(past_days=14)
#print(json.dumps(apiusage, indent=2)) # for development
print(printable_vector_usage(apiusage)) # this needs to be written
if __name__ == "__main__":
logging.basicConfig(level=logging.INFO)
main()
1 Like
Thanks! I ran this code – actually a port into Typescript for my own convenience. It just calls the org/usage/vector_stores REST APIs. Anyway, the output is consistent with what I see in the UI – both prices and usage (bytes).
Then I noticed that if you take the usage numbers for a day in bytes, divide by 2^30 (GB), divide by 24 (hours/day), and multiply by 0.1 ($0.10 per GB/day), you’ll get exactly the corresponding price for that day in the table. So the usage numbers reported are the sum of the hourly usage numbers for the day. That’s not very intuitive, but at least I now understand what I’m looking at.
Thanks for your attention to this! I wasn’t aware of those admin APIs and am happy to know that they’re there when I need them again.
| UTC Date | Line Item | Amount (USD) |
|---|---|---|
| 2025-11-06T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0079 |
| gpt-5-mini-2025-08-07, input | 2.3430 | |
| gpt-5-mini-2025-08-07, cached input | 0.0006 | |
| gpt-5-mini-2025-08-07, cached input | 0.0475 | |
| gpt-5-mini-2025-08-07, output | 0.0058 | |
| gpt-5-mini-2025-08-07, output | 1.0373 | |
| file search storage | 0.5780 | |
| file search tool calls | 0.0025 | |
| file search tool calls | 0.9125 | |
| 2025-11-07T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0028 |
| gpt-5-mini-2025-08-07, input | 1.6016 | |
| gpt-5-mini-2025-08-07, cached input | 0.0156 | |
| gpt-5-mini-2025-08-07, output | 0.0018 | |
| gpt-5-mini-2025-08-07, output | 0.8141 | |
| assistants api | file search | |
| file search storage | 0.5839 | |
| file search tool calls | 0.0050 | |
| file search tool calls | 0.7350 | |
| 2025-11-08T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0337 |
| gpt-5-mini-2025-08-07, input | 1.0512 | |
| gpt-5-mini-2025-08-07, cached input | 0.0003 | |
| gpt-5-mini-2025-08-07, cached input | 0.0078 | |
| gpt-5-mini-2025-08-07, output | 0.0108 | |
| gpt-5-mini-2025-08-07, output | 0.5870 | |
| file search storage | 0.5952 | |
| file search tool calls | 0.0275 | |
| file search tool calls | 0.2450 | |
| 2025-11-09T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.9403 |
| gpt-5-mini-2025-08-07, cached input | 0.0056 | |
| gpt-5-mini-2025-08-07, output | 0.4406 | |
| file search storage | 0.5950 | |
| file search tool calls | 0.2650 | |
| 2025-11-10T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0045 |
| gpt-5-mini-2025-08-07, input | 2.2825 | |
| gpt-5-mini-2025-08-07, cached input | 0.0003 | |
| gpt-5-mini-2025-08-07, cached input | 0.0447 | |
| gpt-5-mini-2025-08-07, output | 0.0058 | |
| gpt-5-mini-2025-08-07, output | 1.1305 | |
| file search storage | 0.5952 | |
| file search tool calls | 0.0025 | |
| file search tool calls | 1.1400 | |
| 2025-11-11T00:00:00Z | gpt-5-mini-2025-08-07, input | 4.5193 |
| gpt-5-mini-2025-08-07, input | 1.8011 | |
| gpt-5-mini-2025-08-07, input | 0.0042 | |
| gpt-5-mini-2025-08-07, cached input | 0.0176 | |
| gpt-5-mini-2025-08-07, cached input | 0.0340 | |
| gpt-5-mini-2025-08-07, output | 4.6655 | |
| gpt-5-mini-2025-08-07, output | 0.8306 | |
| gpt-5-mini-2025-08-07, output | 0.0012 | |
| assistants api | file search | |
| file search storage | 0.5909 | |
| file search tool calls | 0.0100 | |
| file search tool calls | 0.7900 | |
| 2025-11-12T00:00:00Z | gpt-5-mini-2025-08-07, input | 6.3718 |
| gpt-5-mini-2025-08-07, input | 2.1092 | |
| gpt-5-mini-2025-08-07, input | 0.0127 | |
| gpt-5-mini-2025-08-07, cached input | 0.0207 | |
| gpt-5-mini-2025-08-07, cached input | 0.0397 | |
| gpt-5-mini-2025-08-07, output | 4.9934 | |
| gpt-5-mini-2025-08-07, output | 0.9381 | |
| gpt-5-mini-2025-08-07, output | 0.0113 | |
| file search storage | 0.5589 | |
| file search tool calls | 0.0025 | |
| file search tool calls | 0.9675 | |
| 2025-11-13T00:00:00Z | gpt-5-mini-2025-08-07, input | 4.6956 |
| gpt-5-mini-2025-08-07, input | 1.9069 | |
| gpt-5-mini-2025-08-07, cached input | 0.0031 | |
| gpt-5-mini-2025-08-07, cached input | 0.0165 | |
| gpt-5-mini-2025-08-07, output | 3.8464 | |
| gpt-5-mini-2025-08-07, output | 0.8329 | |
| assistants api | file search | |
| file search storage | 0.5551 | |
| file search tool calls | 0.0525 | |
| file search tool calls | 0.7575 | |
| 2025-11-14T00:00:00Z | gpt-5.1-2025-11-13, input | 0.2035 |
| gpt-5.1-2025-11-13, cached input | 0.0207 | |
| gpt-5.1-2025-11-13, output | 0.0196 | |
| gpt-5-mini-2025-08-07, input | 0.0073 | |
| gpt-5-mini-2025-08-07, input | 2.2774 | |
| gpt-5-mini-2025-08-07, input | 0.0045 | |
| gpt-5-mini-2025-08-07, cached input | 0.0006 | |
| gpt-5-mini-2025-08-07, cached input | 0.0479 | |
| gpt-5-mini-2025-08-07, cached input | 0.0000 | |
| gpt-5-mini-2025-08-07, output | 0.0053 | |
| gpt-5-mini-2025-08-07, output | 0.8848 | |
| gpt-5-mini-2025-08-07, output | 0.0047 | |
| file search storage | 0.5577 | |
| file search tool calls | 0.0075 | |
| file search tool calls | 0.7850 | |
| file search tool calls | 0.0050 | |
| 2025-11-15T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0088 |
| gpt-5-mini-2025-08-07, input | 0.8831 | |
| gpt-5-mini-2025-08-07, cached input | 0.0012 | |
| gpt-5-mini-2025-08-07, cached input | 0.0082 | |
| gpt-5-mini-2025-08-07, output | 0.0066 | |
| gpt-5-mini-2025-08-07, output | 0.3635 | |
| file search storage | 0.5535 | |
| file search tool calls | 0.0050 | |
| file search tool calls | 0.2650 | |
| 2025-11-16T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.7399 |
| gpt-5-mini-2025-08-07, cached input | 0.0053 | |
| gpt-5-mini-2025-08-07, output | 0.3136 | |
| file search storage | 0.5603 | |
| file search tool calls | 0.1925 | |
| 2025-11-17T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0112 |
| gpt-5-mini-2025-08-07, input | 2.3530 | |
| gpt-5-mini-2025-08-07, cached input | 0.0002 | |
| gpt-5-mini-2025-08-07, cached input | 0.0361 | |
| gpt-5-mini-2025-08-07, output | 0.0042 | |
| gpt-5-mini-2025-08-07, output | 0.9506 | |
| file search storage | 0.5608 | |
| file search tool calls | 0.0075 | |
| file search tool calls | 0.9125 | |
| 2025-11-18T00:00:00Z | gpt-5-mini-2025-08-07, input | 0.0086 |
| gpt-5-mini-2025-08-07, input | 3.4296 | |
| gpt-5-mini-2025-08-07, cached input | 0.0002 | |
| gpt-5-mini-2025-08-07, cached input | 0.0529 | |
| gpt-5-mini-2025-08-07, output | 0.0030 | |
| gpt-5-mini-2025-08-07, output | 1.3768 | |
| file search storage | 0.5173 | |
| file search tool calls | 0.0025 | |
| file search tool calls | 1.2575 | |
| 2025-11-19T00:00:00Z | gpt-5-mini-2025-08-07, cached input | 0.0003 |
| gpt-5-mini-2025-08-07, cached input | 0.0418 | |
| gpt-5-mini-2025-08-07, input | 0.0028 | |
| gpt-5-mini-2025-08-07, input | 2.8542 | |
| gpt-5-mini-2025-08-07, output | 0.0029 | |
| gpt-5-mini-2025-08-07, output | 1.0665 | |
| file search tool calls | 1.0750 |
| UTC Date | Project ID | Usage (bytes) |
|---|---|---|
| 2025-11-06T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 148953696115 |
| 2025-11-07T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 151324695093 |
| 2025-11-08T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 153381050182 |
| 2025-11-09T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 153323885517 |
| 2025-11-10T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 153383192439 |
| 2025-11-11T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 153134816140 |
| 2025-11-12T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 144020423447 |
| 2025-11-13T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 143916750728 |
| 2025-11-14T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 143718985127 |
| 2025-11-15T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 142628522018 |
| 2025-11-16T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 144393040899 |
| 2025-11-17T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 144510066702 |
| 2025-11-18T00:00:00Z | proj_C7puF5AXoXqVCPrPQYODPolv | 133301886084 |
1 Like
Looks like the costs print parsing broke on line_item type “assistants api | file search” 
Need to replace any pipes (|) in the line item text to not break a markdown table…
| UTC Date | Line Item | Amount (USD) |
|---|---|---|
| 2025-10-23T00:00:00Z | priority | gpt-5-mini-2025-08-07, input |
| priority | gpt-5-mini-2025-08-07, output | |
| file search tool calls | 0.0025 |
| UTC Date | Line Item | Amount (USD) |
| -------------------- | ------------------------------------------- | ------------ |
| 2025-10-23T00:00:00Z | priority | gpt-5-mini-2025-08-07, input | 0.0000 |
| | priority | gpt-5-mini-2025-08-07, output | 0.0000 |
| | file search tool calls | 0.0025 |