feat(ui): show daily token totals on the model leaderboard by ishaan-berri · Pull Request #44044 · BerriAI/litellm

ishaan-berri · 2026-10-01T17:52:04Z

TLDR

Problem this solves:

Model Leaderboard only charts weekly buckets
No way to read actual tokens used on a given day
Tooltip never shows the bucket's total

How it solves it:

Adds a Daily / Weekly toggle to the Top models chart, Daily by default
/model-insights now also returns per-day totals across every model
Tooltip header shows that gateway total, not just the charted top ten

Intentional product change: the Top models chart opens on daily bars instead of weekly ones. Weekly is one click away

User Flow

Before: an admin who wants to know how many tokens the gateway used yesterday can only see 7-day totals

They open http://localhost:4000/ui/?page=model-insights with the Tokens tab selected
The chart says "Weekly tokens across your gateway" and has only Linear / Log toggles
Hovering a bar shows a week start date and per-model counts for the whole week, with no total

Before screenshot (changed areas boxed in red):

After: the same admin sees one bar per day and the day's total token count

They open http://localhost:4000/ui/?page=model-insights with the Tokens tab selected
The chart says "Daily tokens across your gateway" and has a new Daily / Weekly toggle, Daily selected
Hovering a bar shows that day's date, Gateway total 3.1B, and per-model counts for that single day
On a day where only models outside the top ten ran, the header still shows the real total (Gateway total 147) while every charted model reads 0
Clicking Weekly brings back the old weekly bars, also with the gateway total in the tooltip

After screenshot (changed areas boxed in red):

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

I have added meaningful tests
The handful of test files covering my change pass locally, e.g. uv run pytest tests/unit/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
My PR's scope is as isolated as possible; it only solves 1 specific problem
I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Setup: proxy from this branch on localhost:41926 (4000 was taken by another local proxy) (litellm/proxy/dev_config.yaml) against a Postgres that already had a year of per-model daily usage rows, Admin UI dev server on localhost:3001, logged in as the env admin

Before (`91ff045`)

Open http://localhost:3001/model-insights, Tokens tab
The chart shows weekly bars only. Hovering the last full week shows 2026-09-24 and per-model weekly counts with no total (see the Before screenshot in User Flow)

After (`788b6de`)

The API now returns gateway-wide totals. On 2026-10-01 only three models outside the year's top ten ran (two real gpt-4o-mini calls plus earlier failed Anthropic calls), so the top-ten rows are empty but the total is not

$ curl -s "localhost:41926/model-insights?metric=tokens" -H "Authorization: Bearer $KEY" \
    | jq '{top_ten_rows_on_oct_1: [.daily[] | select(.date=="2026-10-01")], gateway_total_on_oct_1: [.daily_totals[] | select(.date=="2026-10-01") | .prompt_tokens + .completion_tokens]}'
{
  "top_ten_rows_on_oct_1": [],
  "gateway_total_on_oct_1": [147]
}

Open http://localhost:3001/model-insights, Tokens tab. The chart shows one bar per day with Daily selected. Hovering 2026-09-20 shows 2026-09-20 · Gateway total 3.1B (see the first After screenshot in User Flow)
Hovering 2026-10-01 shows 2026-10-01 · Gateway total 147 while every charted model reads 0 (see the second After screenshot)
Clicking Weekly switches back to 7-day bars, and the subtitle changes to "Weekly tokens across your gateway"

Type

🆕 New Feature

Caveats (if any)

Low

A full year of daily bars is dense, so bars are thin at the default range
The leaderboard share and delta below the chart are unchanged and still cover the whole period
/model-insights runs one extra grouped query, by date only, over the same bounded window

Final Attestation

The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

🤖 Generated with Claude Code

greptile-apps · 2026-10-01T17:54:54Z

Confidence Score: 4/5

[Medium risk] Adds daily granularity option to the model insights chart.

The PR should not merge until the daily tooltip stops presenting a top-ten subtotal as the gateway total.

Findings

Total excludes other models ▶
Daily bars are too dense ▶
Tooltip total is untested ▶

Summary

The PR adds daily-by-default buckets, a Daily/Weekly chart toggle, and a formatted tooltip total.

Daily and weekly aggregation are covered by data-level tests.
The displayed total excludes models outside the backend’s top-ten selection, and the default year-long daily chart is difficult to inspect.
The tooltip output itself is not asserted by the new view test.

_{Reviews (1) · Last reviewed commit: "test(ui): cover the daily/weekly toggle ..."}

greptile-apps · 2026-10-01T17:54:57Z

+                    labelFormatter={(label, payload) =>
+                      `${label} · Total ${formatMetric(
+                        payload.reduce((sum, item) => sum + Number(item.value ?? 0), 0),
+                        shown,
+                      )}`


Total excludes other models

When a gateway has more than ten models, /model-insights supplies daily rows only for the top ten. This tooltip labels their sum “Total,” so an admin checking yesterday’s gateway token usage sees an undercount. Include the omitted usage or label the figure as the total for the displayed models.

greptile-apps · 2026-10-01T17:54:57Z

  const [loaded, setLoaded] = React.useState<{ metric: Metric; response: ModelInsightsResponse } | null>(null);
  const [metric, setMetric] = React.useState<Metric>("tokens");
  const [scale, setScale] = React.useState<Scale>("linear");
+  const [granularity, setGranularity] = React.useState<Granularity>("day");


Daily bars are too dense

The default response covers 365 days, so opening the chart now fits 365 daily categories into the same width that previously held about 53 weekly categories. At ordinary dashboard widths, the bars become hard to see or hover, making the new default view difficult to use. A shorter daily window or another way to inspect individual days would help.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

greptile-apps · 2026-10-01T17:54:57Z

+  it("charts one bar per day by default and switches to weekly bars", async () => {
+    render(<ModelInsightsView accessToken="token" />);
+    await screen.findByText("fast-chat");
+    const chart = screen.getByTestId("usage-chart");
+    const days = (Date.parse(response.end_date) - Date.parse(response.start_date)) / 86_400_000 + 1;
+
+    expect(screen.getByRole("tab", { name: "Daily" })).toHaveAttribute("aria-selected", "true");
+    expect(chart).toHaveAttribute("data-buckets", String(days));
+    expect(screen.getByText("Daily tokens across your gateway")).toBeInTheDocument();
+
+    await userEvent.click(screen.getByRole("tab", { name: "Weekly" }));
+
+    expect(chart).toHaveAttribute("data-buckets", String(Math.ceil(days / 7)));
+    expect(chart).toHaveAttribute("data-first", response.start_date);
+    expect(screen.getByText("Weekly tokens across your gateway")).toBeInTheDocument();
+  });


Tooltip total is untested

This test checks bucket counts and subtitles, but both tooltip components are mocked to null. A wrong total or metric in the new tooltip would therefore leave the tests green. Assert the displayed date and total using a representative multi-model payload.

…rboard chart

…ust the top ranked ones

…king

…tals

…derboard tooltip

CLAassistant · 2026-10-01T18:14:49Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

codecov · 2026-10-01T18:18:30Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

ishaan-berri · 2026-10-01T18:28:11Z

@greptile approve

greptile-apps · 2026-10-01T18:28:42Z

Approved. The follow-up now computes daily_totals with an unfiltered date-only aggregation, so the tooltip’s gateway total includes models outside the ranked top ten. Daily and weekly bucket alignment is covered by the data tests, and the view test verifies the daily default plus weekly toggle. The remaining density trade-off is intentional and documented in the PR.

codspeed · 2026-10-01T18:29:14Z

Merging this PR will not alter performance

✅ 31 untouched benchmarks

_{Comparing litellm_model_leaderboard_daily_tokens (788b6de) with main (259c166)¹}

No successful run was found on main (a76b59d) during the generation of this report, so 259c166 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

ishaan-berri requested a review from a team October 1, 2026 17:52

greptile-apps Bot reviewed Oct 1, 2026

View reviewed changes

ishaan-berri added 12 commits October 1, 2026 11:01

feat(ui): bucket model leaderboard usage by day or week

2a8aafd

test(ui): cover daily buckets in model leaderboard series

1a3bed9

feat(ui): add daily/weekly toggle and per-bucket total to model leade…

e2e8338

…rboard chart

test(ui): cover the daily/weekly toggle on the model leaderboard

a79fa6a

feat(model-insights): add gateway-wide daily totals to the response type

5e190f0

feat(model-insights): return per-day totals across every model, not j…

4daf6b8

…ust the top ranked ones

test(model-insights): daily totals include models outside the top ran…

5bc87c5

…king

chore(model-insights): regenerate lazy openapi snapshot for daily totals

015e065

chore(ui): regenerate api types for model insights daily totals

6f18d39

fix(ui): compute leaderboard bucket totals from gateway-wide daily to…

afe7d57

…tals

fix(ui): show the gateway total, not the top-ten subtotal, in the lea…

6e2cd23

…derboard tooltip

test(ui): cover gateway-wide bucket totals on the model leaderboard

d6c8b65

ishaan-berri force-pushed the litellm_model_leaderboard_daily_tokens branch from de7ca64 to d6c8b65 Compare October 1, 2026 18:14

ishaan-berri added 2 commits October 1, 2026 11:37

fix(model-insights): type daily totals as an immutable tuple

7582581

fix(model-insights): build daily totals without new mutable collections

788b6de

ishaan-berri enabled auto-merge (squash) October 1, 2026 19:07

moe-berri approved these changes Oct 1, 2026

View reviewed changes

ishaan-berri merged commit fcf8797 into main Oct 1, 2026
98 of 101 checks passed

ishaan-berri deleted the litellm_model_leaderboard_daily_tokens branch October 1, 2026 19:12

mateo-berri mentioned this pull request Oct 1, 2026

chore(lint): remove the LIT002 mutable-construction rule #43971

Merged

6 tasks done

Navigation Menu

Repository navigation

feat(ui): show daily token totals on the model leaderboard - #44044

feat(ui): show daily token totals on the model leaderboard#44044
ishaan-berri merged 14 commits into
mainfrom
litellm_model_leaderboard_daily_tokens

ishaan-berri commented Oct 1, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Oct 1, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot Oct 1, 2026

Uh oh!

greptile-apps Bot Oct 1, 2026

Uh oh!

greptile-apps Bot Oct 1, 2026

Uh oh!

CLAassistant commented Oct 1, 2026

Uh oh!

codecov Bot commented Oct 1, 2026 •

edited

Loading

Uh oh!

ishaan-berri commented Oct 1, 2026

Uh oh!

greptile-apps Bot commented Oct 1, 2026

Uh oh!

codspeed Bot commented Oct 1, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

ishaan-berri commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

TLDR

User Flow

Pre-Submission checklist

Delays in PR merge?

Screenshots / Proof of Fix

Before (91ff045)

After (788b6de)

Type

Caveats (if any)

Low

Final Attestation

Uh oh!

greptile-apps Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Confidence Score: 4/5

Findings

Summary

Uh oh!

greptile-apps Bot Oct 1, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Oct 1, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Oct 1, 2026

Choose a reason for hiding this comment

Uh oh!

CLAassistant commented Oct 1, 2026

Uh oh!

codecov Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

ishaan-berri commented Oct 1, 2026

Uh oh!

greptile-apps Bot commented Oct 1, 2026

Uh oh!

codspeed Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Footnotes

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ishaan-berri commented Oct 1, 2026 •

edited

Loading

Before (`91ff045`)

After (`788b6de`)

greptile-apps Bot commented Oct 1, 2026 •

edited

Loading

codecov Bot commented Oct 1, 2026 •

edited

Loading

codspeed Bot commented Oct 1, 2026 •

edited

Loading