openai/gpt-5.4 limit.input may be using the pricing breakpoint instead of the true max input

[nyl199310]

Problem

providers/openai/models/gpt-5.4.toml currently defines:

[limit]
context = 1_050_000
input = 272_000
output = 128_000

However, the OpenAI docs for GPT-5.4 say the model has a 1.05M context window, and also note:

For models with a 1.05M context window (GPT-5.4 and GPT-5.4 pro), prompts with >272K input tokens are priced at 2x input and 1.5x output for the full session for standard, batch, and flex.

Source: https://developers.openai.com/api/docs/models/gpt-5.4

That wording reads like a pricing threshold, not a hard maximum accepted input size.

Why this matters

Projects consuming models.dev use limit.input as an actual input cap for overflow handling / session compaction. With the current metadata, clients may compact at ~252K input tokens even when the backend can successfully accept much larger requests.

Request

Could you clarify whether 272_000 is intended to be the real max input token limit for openai/gpt-5.4, or whether it is the pricing breakpoint?

If it is only the pricing breakpoint, would it make sense to:

• change limit.input to the true accepted max input size, or
• remove / relax limit.input for this model, and
• represent the >272K pricing behavior separately

If 272_000 is in fact the real hard input limit, could you link the authoritative source for that?