feat: add model_group filter to /spend/logs/v2 endpoint

[silencedoctor]

Relevant issues

Fixes #24781

Pre-Submission checklist

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Type

🆕 New Feature

Changes

Add an optional model_group query parameter to the /spend/logs/v2 and /spend/logs/ui endpoints, enabling users to filter spend logs by model group (the public-facing model name used by the Router for load-balancing).

What changed

litellm/proxy/spend_tracking/spend_management_endpoints.py

• Added model_group as an optional query parameter (consistent with existing model and model_id params)
• Added Prisma where_conditions for model_group filtering
• Added model_group to the raw SQL equality filter list

tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py

• Added test_ui_view_spend_logs_with_model_group following the same pattern as the existing test_ui_view_spend_logs_with_model and test_ui_view_spend_logs_with_model_id tests

Why

The model_group column already exists in the LiteLLM_SpendLogs table and is returned in responses, but there was no way to filter by it. This is needed for downstream consumers that want to analyze usage at the model-group level (e.g., all deployments of "gpt-4") rather than individual deployment models.

Test plan

• test_ui_view_spend_logs_with_model_group — verifies that passing model_group=gpt-4 returns only logs with that model group
• All 51 existing tests in test_spend_management_endpoints.py continue to pass
• Adjacent test_ui_view_spend_logs_with_model and test_ui_view_spend_logs_with_model_id tests unaffected

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Apr 15, 2026 7:55am

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
12 out of 17 committers have signed the CLA.

✅ lucassz
✅ hatim-ez
✅ Ashton-Sidhu
✅ daanhendrio
✅ ryan-crabbe-berri
✅ emerzon
✅ duan-levan
✅ yuneng-berri
✅ Sameerlite
✅ shivamrawat1
✅ silencedoctor
✅ joereyna
❌ krisyang1125
❌ krrish-berri-2
❌ ti3x
❌ ishaan-berri
❌ jonemo
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codspeed-hq]

Congrats! CodSpeed is installed 🎉

🆕 16 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

Detected benchmarks

• test_token_counter_simple_message: 235.8 µs
• test_get_llm_provider_with_prefix: 136.4 µs
• test_get_llm_provider_openai: 137.8 µs
• test_get_model_info_with_provider: 80 µs
• test_get_model_info_anthropic: 79 µs
• test_token_counter_long_content: 1.7 ms
• test_cost_per_token_openai: 540.7 µs
• test_token_counter_multi_turn: 567.7 µs
• test_get_llm_provider_anthropic: 141.3 µs
• test_token_counter_with_tools: 411.7 µs
• test_get_llm_provider_azure: 143.5 µs
• test_cost_per_token_anthropic: 540.6 µs
• test_token_counter_raw_text: 182.4 µs
• test_get_model_info_openai: 80.2 µs
• test_get_model_cost_key_exact_match: 79.7 µs
• test_get_model_cost_key_case_insensitive: 81.8 µs

---

[greptile-apps]

Greptile Summary

This PR adds model_group as an optional query parameter to the /spend/logs/v2 and /spend/logs/ui endpoints, enabling filtering by the Router's public model group name. The implementation follows the exact same pattern as the existing model and model_id filters, correctly wiring model_group into both the Prisma count query and the parameterized raw SQL data query. A corresponding unit test is included and matches the style of adjacent tests.

Confidence Score: 5/5

Safe to merge — the change is minimal, follows existing patterns exactly, and the filter value is properly parameterized in raw SQL to prevent injection.

All findings are P2 or below. The implementation correctly adds the filter to both the Prisma count path and the parameterized raw SQL data path, matching the pattern of model and model_id filters. The test covers the happy path using the established mock helper.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/proxy/spend_tracking/spend_management_endpoints.py	Adds model_group as an optional FastAPI query parameter and correctly propagates it into both the Prisma where_conditions and the parameterized raw SQL equality filter list.
tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py	Adds test_ui_view_spend_logs_with_model_group using the existing make_ui_spend_logs_mock_prisma helper; also reformats several multi-context with statements to Python 3.10+ parenthesized style (no functional change).

Sequence Diagram

sequenceDiagram
    participant Client
    participant FastAPI as /spend/logs/v2 or /spend/logs/ui
    participant Prisma as Prisma Count
    participant SQL as Raw SQL Query

    Client->>FastAPI: GET ?model_group=gpt-4&start_date=...&end_date=...
    FastAPI->>FastAPI: Build where_conditions {model_group: "gpt-4", startTime: {...}}
    FastAPI->>Prisma: litellm_spendlogs.count(where=where_conditions)
    Prisma-->>FastAPI: total_records
    FastAPI->>FastAPI: Build SQL WHERE clause model_group = $N (parameterized)
    FastAPI->>SQL: query_raw(sql, ..., "gpt-4", ...)
    SQL-->>FastAPI: paginated rows
    FastAPI-->>Client: {data, total, page, page_size, total_pages}

Loading

_{Reviews (3): Last reviewed commit: "feat(proxy): add model_group filter to /..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[silencedoctor]

Hi @krrish-berri-2 @ishaan-jaff 👋

This PR has been open for over 2 weeks. I've rebased onto the latest main, fixed the uv.lock sync issue, and applied Black formatting — CI should be green now.

Greptile gave it a 5/5 confidence score and all modified lines are covered by tests.

The change itself is minimal: it adds an optional model_group query parameter to /spend/logs/v2, following the exact same pattern as the existing model and model_id filters.

Use case: We're building an AI Gateway dashboard where end users should only see the logical model name (e.g. "gpt-4"), without needing to know which specific provider deployment is serving their requests behind the scenes. The model_group column already exists in SpendLogs and is returned in responses — this PR just adds the missing filter parameter so we can query logs at the model-group level.

Would appreciate a review when you get a chance. Thanks!

[silencedoctor]

Closing in favor of #26080, which contains the same diff cleanly rebased onto current main. The branch here had drifted significantly behind main and the PR sat un-reviewed for 3 weeks.

New PR: #26080
New issue: #26079