feat(openai): add gpt-5.4-pro to model map

[Sameerlite]

Summary

Adds GPT-5.4 pro to the LiteLLM model map.

Changes

• gpt-5.4-pro – main model
• gpt-5.4-pro-2026-03-05 – snapshot

Model specs

Property	Value
Input	$30/1M tokens
Output	$180/1M tokens
Context window	1,050,000 tokens
Max output tokens	128,000
Priority pricing (>272K input)	2x input, 1.5x output
Reasoning effort	medium, high, xhigh
Endpoints	Chat Completions, Batch, Responses
Modalities	Text, image (input); text (output)
Structured outputs	Not supported

Made with Cursor

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Building	Preview, Comment	Mar 6, 2026 4:23am

[greptile-apps]

Greptile Summary

This PR adds two new OpenAI model entries — gpt-5.4-pro and its dated snapshot gpt-5.4-pro-2026-03-05 — to LiteLLM's model pricing and context window maps. The model specifications (input/output costs, context window, max output tokens, reasoning support, priority pricing multipliers) align with the values described in the PR and follow patterns established by similar models such as azure/o3-pro and azure/o3-deep-research.

Key observations:

• Pricing arithmetic checks out: input $30/1M (3e-05), output $180/1M (1.8e-04), priority input 2× (6e-05), priority output 1.5× (2.7e-04).
• mode: "responses" combined with /v1/chat/completions in supported_endpoints is consistent with precedents like azure/o3-pro (line 4764).
• Both entries omit input_cost_per_token_batches and output_cost_per_token_batches despite declaring /v1/batch as a supported endpoint — this means batch-job cost tracking will be inaccurate.
• Both files (model_prices_and_context_window.json and the backup) carry the same changes, which is expected.

Confidence Score: 3/5

• Safe to merge with minor corrections — missing batch pricing fields should be added before merging to ensure accurate cost tracking for batch requests.
• The change is purely additive (JSON data only), so runtime breakage is unlikely. However, the missing input_cost_per_token_batches and output_cost_per_token_batches fields mean LiteLLM will not correctly report costs for batch-API usage of this model, which is a functional gap given that /v1/batch is advertised as a supported endpoint.
• Both model_prices_and_context_window.json and litellm/model_prices_and_context_window_backup.json need the batch pricing fields added to both gpt-5.4-pro and gpt-5.4-pro-2026-03-05.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds gpt-5.4-pro and gpt-5.4-pro-2026-03-05 model entries. Pricing and context window values look correct, but both entries are missing input_cost_per_token_batches and output_cost_per_token_batches despite listing /v1/batch as a supported endpoint.
litellm/model_prices_and_context_window_backup.json	Backup copy of the main pricing file — identical changes with the same missing batch pricing fields issue.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Caller invokes gpt-5.4-pro] --> B{Endpoint?}
    B -->|/v1/chat/completions| C[responses mode handler]
    B -->|/v1/responses| C
    B -->|/v1/batch| D[Batch handler]
    C --> E{Priority tier?}
    E -->|input ≤ 272K tokens| F[Standard pricing\n$30 input / $180 output per 1M]
    E -->|input > 272K tokens| G[Priority pricing\n$60 input / $270 output per 1M]
    D --> H[Batch pricing\n⚠ Missing: fields not set\ninput_cost_per_token_batches\noutput_cost_per_token_batches]

Loading

_{Last reviewed commit: 0505034}

[greptile-apps]

Missing batch pricing fields

The model lists /v1/batch in supported_endpoints but omits input_cost_per_token_batches and output_cost_per_token_batches. All other models in this file that expose the batch endpoint include these fields — for example, gpt-5-pro at line 21123–21130 has input_cost_per_token_batches: 7.5e-06 and output_cost_per_token_batches: 6e-05, and azure/o3-pro (lines 4759–4766) has both batch cost fields alongside mode: "responses" and /v1/batch support.

Without these fields, LiteLLM cannot accurately track or report costs for batch requests submitted to this model. Based on OpenAI's typical 50% batch discount, the missing values would be:

• input_cost_per_token_batches: 1.5e-05 (i.e. $15/1M)
• output_cost_per_token_batches: 9e-05 (i.e. $90/1M)

The same omission applies to the gpt-5.4-pro-2026-03-05 snapshot entry at line 21082.

[mreza-kiani]

Hey @Sameerlite, I raised this issue: #23014 about gpt-5.4-pro not supporting the completion API. Could you please double-check whether it’s valid? Also, is there anything I can do to help resolve it? Thank you