Add gemma-4-26b-a4b-it-maas to google provider (Google MaaS API model ID)

[awg66]

Summary

The Google Gemini Enterprise Agent Platform (MaaS) uses gemma-4-26b-a4b-it-maas as the callable model ID for Gemma 4 26B A4B IT — the -maas suffix is part of the official GCP API name, not an alias. The current entry in providers/google/models/gemma-4-26b-a4b-it.toml uses the base model name and also has an incorrect max_output value.

Two changes are needed:

1. Add gemma-4-26b-a4b-it-maas as a new entry

providers/google/models/gemma-4-26b-a4b-it-maas.toml — spec from the GCP MaaS model card:

name = "Gemma 4 26B A4B IT (MaaS)"
family = "gemma-4"
release_date = "2026-04-03"
last_updated = "2026-04-03"
attachment = true
reasoning = false
temperature = true
tool_call = true
open_weights = true

[limit]
context = 262_144
output = 128_000

[modalities]
input = ["text", "image"]
output = ["text"]

2. Fix max_output in existing gemma-4-26b-a4b-it.toml

The existing entry has output = 32_768 but the official GCP docs list 128,000 max output tokens for this model. Please correct it.

Source

• GCP MaaS model page
• Model ID: gemma-4-26b-a4b-it-maas
• Context: 262,144 tokens
• Max output: 128,000 tokens
• Inputs: Text, Documents, Images
• Outputs: Text
• Function calling: supported
• Structured output: supported

Impact

Tools that derive LLM model metadata from the google provider (e.g. langchain-google-genai, which runs an automated refresh from models.dev) will not find a ModelProfile for gemma-4-26b-a4b-it-maas when that model ID is used directly. Downstream consumers then fall back to conservative hardcoded limits rather than the model's actual context window.