fix(models): drop retired xAI models and mark multi-agent responses-only

[msdianprince-7]

TLDR

Problem this solves:

• The registry advertises xAI models that no longer exist
• Picking one gets a 404 from xAI, not a redirect
• Multi-agent models are listed as chat models
• xAI does not serve them on Chat Completions at all

How it solves it:

• Remove the grok-2 family, grok-beta and grok-vision-beta
• Mark both multi-agent models responses-only
• Leave every model xAI still serves untouched

User Flow

Before: a developer browsing the gateway's xAI models picks one that xAI stopped serving, and only finds out at request time

1. They open https://litellm-domain/ui/?page=models and see xai/grok-2, xai/grok-2-vision, xai/grok-beta and xai/grok-vision-beta offered with full pricing and a 131k context window
2. They add xai/grok-2 to their config and send POST https://litellm-domain/v1/chat/completions
3. The call fails with a 404 from xAI saying the model does not exist, and no amount of config fixes it
4. They try xai/grok-4.20-multi-agent-0309, which the same page lists as a chat model, on POST https://litellm-domain/v1/chat/completions
5. That fails too, because xAI serves the multi-agent models only on the Responses API

After: the list only offers models xAI still serves, and the multi-agent models are pointed at the right endpoint

1. They open https://litellm-domain/ui/?page=models and the retired slugs are gone, so there is nothing dead to pick
2. xai/grok-4.20-multi-agent-0309 now shows as a /v1/responses model rather than a chat model
3. They send it to POST https://litellm-domain/v1/responses and it works
4. xai/grok-4.3, xai/grok-4.6 and every other live xAI model are listed and priced exactly as before

Relevant issues

Fixes #38179

Linear ticket

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/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).

Sources

• https://docs.x.ai/docs/models no longer lists the grok-2 family, grok-beta or grok-vision-beta
• https://docs.x.ai/developers/migration/may-15-retirement lists the eight slugs that were retired with a redirect to grok-4.3, and the models removed here are not among them, which is why they 404 instead of redirecting
• https://docs.x.ai/developers/model-capabilities/text/multi-agent states "The multi-agent model does not work with the OpenAI Chat Completions API"

Screenshots / Proof of Fix

This is a registry-metadata change, so the user-visible surface is what the proxy reports for a configured deployment. Live proxy on localhost:4000, config:

model_list:
  - model_name: grok-2
    litellm_params: {model: xai/grok-2, api_key: fake-xai-key}
  - model_name: grok-beta
    litellm_params: {model: xai/grok-beta, api_key: fake-xai-key}
  - model_name: grok-multi-agent
    litellm_params: {model: xai/grok-4.20-multi-agent-0309, api_key: fake-xai-key}
  - model_name: grok-4-3
    litellm_params: {model: xai/grok-4.3, api_key: fake-xai-key}

general_settings:
  master_key: sk-qa-38179

Each case reads GET /model/info and prints the resolved model_info for one deployment:

curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'

Before (cd63c7e)

retired slugs are priced as live chat models

1. curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'
2. xai/grok-2: mode=chat input_cost_per_token=2e-06
3. xai/grok-beta: mode=chat input_cost_per_token=5e-06

multi-agent advertised for chat completions

1. curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'
2. xai/grok-4.20-multi-agent-0309: mode=chat supported_endpoints=None

a live xAI model

1. curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'
2. xai/grok-4.3: mode=chat input_cost_per_token=1.25e-06

After (7a65f3a)

retired slugs are priced as live chat models

1. curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'
2. xai/grok-2: mode=None input_cost_per_token=0
3. xai/grok-beta: mode=None input_cost_per_token=0
4. The registry no longer claims these are live chat models

multi-agent advertised for chat completions

1. curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'
2. xai/grok-4.20-multi-agent-0309: mode=responses supported_endpoints=['/v1/responses']

a live xAI model

1. curl -s http://localhost:4000/model/info -H 'Authorization: Bearer sk-qa-38179'
2. xai/grok-4.3: mode=chat input_cost_per_token=1.25e-06, unchanged

No request is sent to xAI here, since I have no xAI key and the retired slugs would only return the 404 the issue already reports.

Type

🐛 Bug Fix

Caveats (if any)

Medium

• Anyone still naming a removed slug loses its cost-map entry, so spend logs for it read 0
  • The call itself already fails at xAI, so there is no real spend to lose
  • Reinstating an entry is the fix if xAI ever brings a slug back

Low

• Removed rather than dated, because xAI published no retirement date for these
  • This repo reverts unverified deprecation dates, see 30b1459
  • xai/grok-2-vision-1212 did already carry deprecation_date: 2026-02-28; if maintainers would rather mark the family than remove it, that is the date to apply
• litellm/constants.py still lists Clarifai-hosted grok-2 slugs, deliberately left alone since Clarifai serves those independently of xAI
• Two older tests still use retired slugs as fixture strings; they exercise provider routing and param mapping, never the cost map, so they are untouched and still pass

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

[greptile-apps]

Greptile Summary

The PR removes eight retired xAI model entries from both model registries and marks two xAI multi-agent models as Responses-only. It adds regression coverage for removals, endpoint metadata, surviving chat models, and registry synchronization.

Confidence Score: 4/5

The PR appears safe to merge, with only non-blocking typing and comment-policy cleanup needed in the new tests.

The xAI Responses path has a registered native adapter and both registries remain synchronized; the only accepted concern is imprecise test typing and unnecessary explanatory comments.

Files Needing Attention: tests/test_litellm/llms/xai/test_xai_model_registry.py

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Removes retired xAI entries and correctly aligns both multi-agent entries with the native Responses endpoint.
litellm/model_prices_and_context_window_backup.json	Mirrors the primary registry changes without xAI metadata drift.
tests/test_litellm/llms/xai/test_xai_model_registry.py	Adds meaningful registry regression coverage but violates repository typing and comment conventions.

_{Reviews (1): Last reviewed commit: "fix(models): drop retired xAI models and..." | Re-trigger Greptile}

[greptile-apps]

Coarse test annotations

The new fixture and three tests use bare dict annotations, reducing static type checking, while the explanatory comments on the model tuples fall outside the repository's source-comment convention. Please use precise map types and remove the unnecessary comments.

Context Used: CLAUDE.md (source)

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!

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[devin-ai-integration]

Closing in favor of the rolling registry PR #38207, which now removes these retired xAI entries and carries the multi-agent Responses-only metadata and registry test

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing msdianprince-7:xai_retire_grok_2_models (7a65f3a) with litellm_internal_staging (8ebcb3e)¹}

Footnotes

1. No successful run was found on litellm_internal_staging (cd63c7e) during the generation of this report, so 8ebcb3e was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩