-
Notifications
You must be signed in to change notification settings - Fork 7.1k
Description
Feature Description
The gpt-5.3 model family (released by OpenAI on March 3, 2026) is missing from the model context size map in llama-index-llms-openai. This causes a ValueError when trying to use any gpt-5.3 variant
The model map in utils.py jumps from gpt-5.2 directly to gpt-5.4, skipping the entire gpt-5.3 family.
Models to add
Based on the OpenAI models page, the following gpt-5.3 models should be added:
gpt-5.3-chat-latest(128K context window)
And any other gpt-5.3 variants available (e.g. gpt-5.3, gpt-5.3-codex, etc.).
Affected file
llama-index-integrations/llms/llama-index-llms-openai/llama_index/llms/openai/utils.py
Specifically the OPENAI_REASONING_MODELS, CHAT_MODELS, and ALL_AVAILABLE_MODELS dictionaries.
Version
Tested up to llama-index-llms-openai==0.7.3 — none of the released versions include gpt-5.3.
Reason
The model context size map in utils.py skips gpt-5.3 entirely — it goes from gpt-5.2 to gpt-5.4. Any attempt to use a gpt-5.3 variant raises ValueError: Unknown model 'gpt-5.3-chat-latest'. Tested all versions up to llama-index-llms-openai==0.7.3, none include it. The only workaround is monkey-patching ALL_AVAILABLE_MODELS at runtime.
Value of Feature
GPT-5.3 Chat is a production OpenAI model available since March 3, 2026. Without it in the model map, users cannot use gpt-5.3-chat-latest or other 5.3 variants through LlamaIndex without monkey-patching. This is a simple addition to the existing model dictionaries in utils.py — the model has a 128K context window, consistent with other chat-latest variants already in the map.