Skip to content

fix: add Bedrock models nvidia.nemotron-super-3-120b#632

Open
github-actions[bot] wants to merge 1 commit into
mainfrom
chore/autofix-issue-624
Open

fix: add Bedrock models nvidia.nemotron-super-3-120b#632
github-actions[bot] wants to merge 1 commit into
mainfrom
chore/autofix-issue-624

Conversation

@github-actions
Copy link
Copy Markdown
Contributor

fix: add Bedrock models nvidia.nemotron-super-3-120b

Closes #624

Source issue: #624

Summary

Field Value
Provider bedrock
Primary model nvidia.nemotron-super-3-120b
Changed models nvidia.nemotron-super-3-120b
Added models nvidia.nemotron-super-3-120b
Updated models None
Verification sources 1
2
3
4

Verified metadata

Model Display name Parent Providers Format Flavor Token limits Pricing Lifecycle
nvidia.nemotron-super-3-120b NVIDIA Nemotron 3 Super 120B bedrock converse chat input=256000, output=32768 n/a active

Verification notes

Verification

Sources and fields verified

Source Fields verified
AWS Bedrock model card (source 1) model ID, format (converse API), context window (256K), max output (32K), modalities (text-only), lifecycle (Active), launch date (2026-03-11), regional availability (12 regions), service tiers
AWS models-at-a-glance (source 2) model listed under NVIDIA provider, model card link
NVIDIA build page (source 4) max_output_tokens = 32,768, model architecture (hybrid Mamba-Transformer MoE, 120B params / 12B active)

Flavor correction

The original issue listed flavor as completion. This has been corrected to chat based on:

  • The model uses the Bedrock Converse API (chat-oriented)
  • All other Bedrock Converse models in model_list.json use flavor: "chat" (e.g., amazon.nova-pro-v1:0, deepseek.v3.2)
  • The model accepts chat-style messages with role/content structure per the AWS sample code

Fields not published or not applicable

  • Pricing (input_cost_per_mil_tokens, output_cost_per_mil_tokens): The Bedrock pricing page lists NVIDIA in the provider menu but the NVIDIA pricing section was not extractable from web fetches due to page truncation. Pricing is omitted rather than guessed.
  • Cache pricing: Not published for this model.
  • parent: Not applicable — this is the base model ID with no stable alias or snapshot relationship.
  • supported_regions: Not applicable — existing Bedrock entries in model_list.json do not use this field (it is used only for Vertex models).
  • reasoning: NVIDIA's own deployment supports reasoning modes, but the Bedrock model card does not document reasoning/thinking token support via the Converse API. Omitted to avoid claiming unverified capability.

sync_models (LiteLLM) cross-check

The model nvidia.nemotron-super-3-120b is not present in the LiteLLM model_prices_and_context_window_backup.json catalog. No bedrock key matching nvidia.nemotron-super-3-120b or bedrock/nvidia.nemotron-super-3-120b was found. Therefore no numeric field comparison is possible. All proposed values are sourced directly from the official AWS Bedrock model card.

Verification checklist

  • Model ID confirmed in >=2 independent official sources (AWS model card + models-at-a-glance)
  • Model ID format matches Bedrock convention (provider.model-name)
  • Not already present in model_list.json (confirmed via grep)
  • Standard/on-demand tier confirmed (not custom-deployment-only)
  • Flavor corrected from completion to chat per Bedrock Converse conventions
  • Token limits verified against both AWS and NVIDIA sources
  • Pricing not verified (Bedrock pricing page NVIDIA section not extractable)

sync_models vs proposed update

sync_models cross-check found differences. Official provider verification was used for the applied values, and sync_models discrepancies are listed below for review.

Model Field Proposed update sync_models sync_models source models
nvidia.nemotron-super-3-120b input_cost_per_mil_tokens n/a 0.15 nvidia.nemotron-super-3-120b
nvidia.nemotron-super-3-120b output_cost_per_mil_tokens n/a 0.65 nvidia.nemotron-super-3-120b

@vercel
Copy link
Copy Markdown

vercel Bot commented May 22, 2026

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
ai-proxy Ready Ready Preview, Comment May 22, 2026 1:21pm

Request Review

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[BOT ISSUE] Bedrock: add missing NVIDIA Nemotron 3 Super 120B model

1 participant