Add amazon.nova-premier-v1:0 to amazon-bedrock provider

[awg66]

Summary

amazon.nova-premier-v1:0 is missing from providers/amazon-bedrock/models/, even though the other Nova generation-1 models (nova-lite, nova-micro, nova-pro) are present there.

The Kilo provider already has a nova-premier entry, but the amazon-bedrock provider is the one consumed by downstream tooling (e.g. langchain-aws runs a daily refresh from models.dev and uses the amazon-bedrock provider).

Spec (from AWS Bedrock docs)

• Model ID: amazon.nova-premier-v1:0
• Geo inference ID: us.amazon.nova-premier-v1:0
• Release date: 2025-10-31
• Knowledge cutoff: 2024-10
• Context window: 1,000,000 tokens
• Max output tokens: 25,000
• Reasoning: supported
• Input modalities: text, image, video
• Output modalities: text
• Tool use: yes
• Prompt caching: yes

Suggested file

providers/amazon-bedrock/models/amazon.nova-premier-v1:0.toml — modeled after amazon.nova-pro-v1:0.toml:

name = "Nova Premier"
family = "nova-premier"
release_date = "2025-10-31"
last_updated = "2025-10-31"
attachment = true
reasoning = true
temperature = true
knowledge = "2024-10"
tool_call = true
open_weights = false

[cost]
# Please verify against https://aws.amazon.com/bedrock/pricing/
input = 2.50
output = 12.50
cache_read = 0.625

[limit]
context = 1_000_000
output = 25_000

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

Impact

Tools that derive LLM model metadata from the amazon-bedrock provider will not emit a ModelProfile for nova-premier. For example, langchain-aws runs a daily automated refresh from models.dev — without an entry here, downstream consumers fall back to conservative hardcoded token limits rather than the model's actual 1M-token context window.