Litellm hotfix azure ai gpt 5.4 main#28104
Conversation
Register azure_ai GPT-5.4 variants with pricing, context limits from Foundry catalog, and capability flags for cost routing and tooling. Co-authored-by: Cursor <cursoragent@cursor.com>
…data Add supports_web_search for base GPT-5.4 aliases, priority-tier Pro rates, and mini/nano above-272k plus priority pricing for correct spend math. Co-authored-by: Cursor <cursoragent@cursor.com>
…ckup row Mirror supports_web_search for azure_ai/gpt-5.4-2026-03-05 in the backup catalog so it matches model_prices_and_context_window.json. Co-authored-by: Cursor <cursoragent@cursor.com>
Greptile SummaryThis PR adds eight new
Confidence Score: 4/5Safe to merge with a quick confirmation that gpt-5.4-pro is intentionally responses-API-only and that the pricing figures match the Azure Foundry catalog. The change is purely additive JSON config — no runtime logic is touched. The pricing math checks out internally, and both files are kept in sync. The two open questions are whether the gpt-5.4-pro chat-completions omission is intentional, and whether the pricing figures have been cross-checked against the official Azure catalog, neither of which can be verified from the PR description alone. The azure_ai/gpt-5.4-pro entry in both JSON files deserves a second look to confirm the responses-only mode and pricing are accurate per the Azure Foundry listing.
|
| Filename | Overview |
|---|---|
| model_prices_and_context_window.json | Adds 8 new azure_ai/gpt-5.4 model variants (base, pro, mini, nano — each with a dated snapshot alias); pricing math is internally consistent; gpt-5.4-pro is responses-mode only (no /v1/chat/completions), which diverges from the other family members and may be intentional but is unverified in the PR description. |
| litellm/model_prices_and_context_window_backup.json | Backup file updated in lock-step with the primary; identical entries and the same internal pricing consistency; same gpt-5.4-pro responses-only concern applies. |
Comments Outside Diff (2)
-
model_prices_and_context_window.json, line 488 (link)gpt-5.4-promissing/v1/chat/completionsendpointazure_ai/gpt-5.4-pro(and its dated variant) set"mode": "responses"and list only/v1/batchand/v1/responsesassupported_endpoints, omitting/v1/chat/completions. The other gpt-5.4 family members all include chat completions. If this model actually supports chat completions on Azure Foundry, this entry will silently mis-represent its capabilities and may confuse users who try the standard chat path. If the omission is intentional (responses-API-only model), a brief comment or source link clarifying that would help reviewers verify the pricing data. -
model_prices_and_context_window.json, line 394 (link)PR description lacks evidence of what was fixed
The PR title says "hotfix" but the description contains only "gpt-5.4 azure_ai for foundry" with no reference to a GitHub issue, no prior bug report, and no test output confirming the entries are correct. Per the team's review standard, PRs claiming to fix something should include verifiable evidence (e.g., a link to the Azure pricing page, a test run, or an issue number). Without a source cross-reference it's hard to confirm that the pricing values and capability flags are accurate.
Rule Used: What: Ensure that any PR claiming to fix an issue ... (source)
Reviews (1): Last reviewed commit: "fix(model_catalog): sync web_search flag..." | Re-trigger Greptile
1 participant
gpt-5.4 azure_ai for foundry