[BOT ISSUE] fix: add missing Bedrock model nvidia.nemotron-super-3-120b (NVIDIA Nemotron 3 Super)

[github-actions]

Missing model

The model nvidia.nemotron-super-3-120b (NVIDIA Nemotron 3 Super 120B) is available on Amazon Bedrock but absent from the model catalog (packages/proxy/schema/model_list.json). The catalog has no Bedrock-format NVIDIA entries at all.

NVIDIA Nemotron 3 Super is a 120B-parameter open hybrid MoE model that activates just 12B parameters, delivering high throughput for agentic tasks. It launched on Bedrock in March 2026, is actively listed with no end-of-life date, and supports Standard, Priority, and Flex service tiers.

Model details

Field	Value
Model ID	nvidia.nemotron-super-3-120b
Format	converse
Flavor	chat
Display name	NVIDIA Nemotron 3 Super 120B
Max input tokens	256000
Max output tokens	32000
Available providers	["bedrock"]

Verification checklist

• Cross-source confirmation: Model confirmed on (1) AWS Bedrock model card with model ID, context window, and max output tokens, (2) AWS "What's New" announcement confirming GA availability (March 2026), and (3) AWS blog post with technical details.
• Recent commits check: No nvidia.* Bedrock-format entries exist in model_list.json.
• ID format validation: Follows Bedrock naming convention (provider.model-name), consistent with existing entries. The model card's sample code uses nvidia.nemotron-super-3-120b for both Invoke API and Chat Completions API calls.

Verification notes

Field	Source
Model ID (nvidia.nemotron-super-3-120b)	Bedrock model card — Programmatic Access table
Context window (256K), max output (32K)	Bedrock model card — Model Details section
GA availability (March 2026)	AWS What's New
Technical details	AWS blog post

Fields NOT verified from official sources:

• input_cost_per_mil_tokens / output_cost_per_mil_tokens — Pricing was not extractable from the Bedrock pricing page. The downstream fix job should retrieve the US standard tier pricing directly from the pricing page's NVIDIA section.
• multimodal — The model card shows text-only input/output. Not multimodal.
• A GovCloud geo inference ID us-gov.nvidia.nemotron-super-3-120b is also documented. The downstream fix job should consider whether a separate entry is needed.

Regional availability

Available in-region in: us-east-1, us-east-2, us-west-2, eu-south-1, eu-west-1, eu-west-2, ap-northeast-1, ap-south-1, ap-southeast-2, sa-east-1, ap-southeast-4, us-gov-west-1.

{
  "kind": "missing_model",
  "provider": "bedrock",
  "models": ["nvidia.nemotron-super-3-120b"],
  "status": "active",
  "model_specs": {
    "nvidia.nemotron-super-3-120b": {
      "format": "converse",
      "flavor": "chat",
      "displayName": "NVIDIA Nemotron 3 Super 120B",
      "max_input_tokens": 256000,
      "max_output_tokens": 32000,
      "available_providers": ["bedrock"]
    }
  },
  "source_urls": [
    "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-nvidia-nemotron-super-3-120b.html",
    "https://aws.amazon.com/about-aws/whats-new/2026/03/amazon-bedrock-nemotron-3-super/",
    "https://aws.amazon.com/blogs/machine-learning/run-nvidia-nemotron-3-super-on-amazon-bedrock/"
  ]
}