fix(model_prices): add azure_ai gpt-image-2 and groq llama-guard-3-8b deprecation dates

[devin-ai-integration]

TLDR

Problem this solves:

• azure_ai/gpt-image-2 has no deprecation_date while azure/gpt-image-2 does
• groq/llama-guard-3-8b was retired by Groq on 2025-06-06 but carries no date

How it solves it:

• Adds deprecation_date to both entries from the provider retirement tables
• Keeps the bundled backup byte-identical and the schema in sync

User Flow

Before: an operator reading /model/info cannot tell that these two models are scheduled for or already past retirement

1. They send GET https://litellm-domain/v1/model/info for azure_ai/gpt-image-2 and groq/llama-guard-3-8b
2. Neither entry has a deprecation_date, so their deprecation dashboard (issue feat(proxy): proactive model deprecation alerts and /model/deprecations endpoint #26900) lists them as open-ended
3. Groq's deprecation page says llama-guard-3-8b was retired on 2025-06-06, but nothing in the registry reflects that

After: both entries carry the provider's published retirement date

1. They send the same GET https://litellm-domain/v1/model/info
2. azure_ai/gpt-image-2 shows "deprecation_date": "2027-10-21" and groq/llama-guard-3-8b shows "deprecation_date": "2025-06-06"
3. Their dashboard flags the Groq model as already retired and schedules the Azure one for October 2027

Relevant issues

Supports #26900 (deprecation metadata in the registry)

Affected release

Linear ticket

Pre-Submission checklist

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

• I have added meaningful tests (data-only change; existing tests/test_litellm/test_model_prices_schema.py and test_cost_map_guard.py validate the two entries against the schema)
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<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

This is the rolling registry PR for the 2026-09-23 audit run. The previous rolling PR #41964 merged on 2026-09-19, so this one is new

Sources for every changed value

Both rows were read from the raw HTML tables (curl with a cache buster, <tr>/<td> parsed), then cross-checked with r.jina.ai

Registry key	Field	Value	Raw provider row	Source
azure_ai/gpt-image-2	deprecation_date	2027-10-21	gpt-image-2 | 2026-04-21 | GA | 2027-10-21 | —	https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule
groq/llama-guard-3-8b	deprecation_date	2025-06-06	llama-guard-3-8b | 06/06/25 | meta-llama/llama-guard-4-12b	https://console.groq.com/docs/deprecations

azure/gpt-image-2 and azure/gpt-image-2-2026-04-21 already carry 2027-10-21 from the same Azure row; the azure_ai/ key is the same Foundry model

Checked and intentionally not changed

• Anthropic: 14 models listed as Active | N/A | Not sooner than <date> on https://docs.anthropic.com/en/docs/about-claude/model-deprecations. Those are earliest-possible retirement floors, not announced dates, so no deprecation_date is written (same reading as fix(model_prices): drop anthropic deprecation floors and correct azure gpt-4.1-nano retirement date #41964)
• Azure gpt-realtime-mini (both versions), gpt-4o-mini-transcribe and gpt-4o-mini-tts undated aliases: the schedule lists two retirement dates per version, so the registry key cannot be mapped to one row
• OpenAI gpt-4o-audio-preview-*, gpt-4o-mini-audio-preview-2024-12-17, gpt-4o-mini-realtime-preview-2024-12-17 keep 2027-01-20: the page lists the undated -preview aliases at 2026-05-07 and gpt-4o-mini-realtime at Jan 20, 2027, without naming these snapshots, so the existing value is left alone
• Pricing spot-check against raw OpenAI, Anthropic, Gemini, Vertex, Bedrock, Azure, Groq, Mistral, Cohere, DeepSeek, xAI, Fireworks tables found zero new mismatches

Open registry PRs reviewed for absorption (none absorbed)

• fix(prices): add estimated Bedrock prices for GPT-6 Sol and GPT-6 Luna #42728 Bedrock GPT-6 Sol/Luna: prices are declared as estimates by the author, not published
• fix(prices): add missing bedrock_mantle entries for DeepSeek, Kimi and Qwen models #42727 Bedrock Mantle DeepSeek/Kimi/Qwen: the base DeepSeek V3.1 and Qwen3 235B prices render as JS placeholders in the raw AWS pricing HTML, and the Kimi K2 Thinking limits (256000/16000) contradict the existing bedrock/moonshotai.kimi-k2-thinking entries (262144/262144), so nothing could be confirmed; also ships a test file, left to its author
• feat(azure): price GPT-6 Sol and GPT-6 Luna on the azure route #42704 Azure GPT-6 Sol/Luna: draft; the Azure pricing page is a JS shell with no parseable rows
• fix(model_prices_and_context_window.json): set regional pricing for b… #42698 applies a 10 percent regional uplift to the unprefixed anthropic.* Bedrock keys, which are the global rows; regional prices belong on region-prefixed keys
• fix(prices): fix xai/grok-4 capabilities #41376 xAI grok-4 capability flags: xAI model docs are JS-rendered, no raw table to verify against
• feat(models): sync together_ai model registry #41775 together_ai sync: maintainer-authored, left alone
• chore(prices): sync Azure prices: 20 models, 20 new [20 with gaps] #42594, chore(prices): sync OpenRouter prices: 19 models, 9 new [18 held] #42592, chore(prices): sync Fireworks AI prices: 2 models, 2 new [2 with gaps] #42590 provider-info-sync bot PRs: bot-owned sync PRs with declared gaps/held rows, left to the sync process
• fix(azure_ai): bill FLUX.2-flex edit reference megapixels at Azure's rate #42448, feat(audio): add Gemini transcription vocabulary support #42547, feat(provider): add Yolo-Auto (yolo-auto) OpenAI-compatible provider #42545, feat(sarvam): add realtime speech to text on /v1/realtime #42479, feat(providers): add uRun as a JSON-configured OpenAI-compatible provider #42343, fix(deepseek): promote reasoning_content on multi-turn default thinking models #42158, feat(search): add Serply as a search provider #42150, feat(providers): add TopxAI as a JSON-configured OpenAI-compatible provider #41919, feat(providers): add Prism provider (internal copy of #40914) #41961, feat(router): add fusion models #39224 and the remaining registry-touching PRs are code or new-provider changes that only incidentally touch the JSON, left alone

Local checks at the PR tip

$ sha256sum model_prices_and_context_window.json litellm/model_prices_and_context_window_backup.json
f2ecfbe8e2e04c483b5e8d2eada912f54e6393d535cc3c3c7805adf32ab1248a  model_prices_and_context_window.json
f2ecfbe8e2e04c483b5e8d2eada912f54e6393d535cc3c3c7805adf32ab1248a  litellm/model_prices_and_context_window_backup.json
$ python3 ci_cd/check_files_match.py
Passed! Files model_prices_and_context_window.json and litellm/model_prices_and_context_window_backup.json match.
$ python ci_cd/generate_model_prices_schema.py --check
model_prices_and_context_window.schema.json is in sync and model_prices_and_context_window.json validates against it
$ python -m pytest tests/test_litellm/test_model_prices_schema.py tests/test_litellm/test_cost_map_guard.py -q
85 passed in 5.48s

Link to Devin session: https://app.devin.ai/sessions/acde7a8280134b8a8e5a16277790d661
Open in Devin Desktop: https://app.devin.ai/desktop/session/acde7a8280134b8a8e5a16277790d661?variant=devin

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge because both dates map to the intended models and satisfy the existing registry contract

Summary

This PR adds provider-published deprecation dates for two existing model registry entries and mirrors both changes in the bundled backup

• Adds the October 21, 2027 retirement date to azure_ai/gpt-image-2
• Adds the June 6, 2025 retirement date to groq/llama-guard-3-8b
• Preserves registry schema compatibility and byte-for-byte synchronization between the primary and backup files

_{Reviews (1) · Last reviewed commit: "fix(model_prices): add azure_ai gpt-imag..."}

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_registry_deprecations_20260923 (6ac5b85) with main (860bc78)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[CLAassistant]

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.}