[Vidit Ostwal (Vidit-Ostwal)]

Note

Low Risk
Data-only catalog sync with no application logic changes; incorrect entries could only affect model resolution or cost estimates.

Overview
Bumps the vendored LiteLLM model_prices_and_context_window.json to v1.15.41, refreshing pricing metadata and model entries used for context windows and cost calculation.

Aliases: Adds *-latest / seedance-latest entries for fal_ai, runwayml, replicate, gemini, and vertex_ai that point at pinned 2.1 / 2.5 model keys.

New models: Registers fal_ai nano-banana-2.1 (generate + edit), Seedream v5.5, Flux VTO, Patina material pipelines, and Vidu template-to-video; runwayml/kling_4; gemini/gemini-decision-1 (/v1/decisions); gemini-4-hydrogen; and gmi gemini-4-carbon / grok4.7.

Pricing / lifecycle updates: Revises Databricks Claude Fable 5.2 rows to DBU-derived dollar rates (including cache tiers); adjusts xAI grok-imagine-image-quality family to $0.04 with 1K/2K output costs; adds Fireworks deprecation sources and successor_model for DeepSeek v4-flash; extends Gemini image token batch/priority costs.

^{Reviewed by Cursor Bugbot for commit 77d4ba0. Bugbot is set up for automated code reviews on this repo. Configure here.}

[strix-security]

Strix Security Review

✅ No open security findings.

PR summary: Updates the bundled LiteLLM model catalog: adds `*-latest` alias entries, new fal_ai, runwayml, gmi and Gemini model rows, deprecation/successor metadata, and revised xAI and Databricks pricing. No application code changes.

Review summary

This PR syncs the vendored LiteLLM model catalog: new alias_of rows for *-latest model names, new fal_ai/runwayml/gmi/gemini entries, added deprecation_source/successor_model fields, and updated xAI and Databricks pricing. The file is data only, and its sole runtime consumer is the context-window lookup in lib/crewai/src/crewai/llms/context_window.py, which reads just max_input_tokens and litellm_provider; every new field (aliases, mode, source, supported_endpoints, cost figures) is never read, and no existing model's resolved window changes. I verified the file still parses with no duplicate keys, that all new alias targets exist and resolve in one hop with no cycles, and that a user-supplied model name only reaches a dictionary lookup with a safe default fallback, so no injection, SSRF, key-routing or control-weakening path exists here. No security issues found in the change.

_{Review 1 on this PR · 77d4ba0 · requested Nov 12 18:23 UTC when the PR was opened}

---

Reviewed by Strix
_{Re-run review · Configure security review settings}