feat(aws-bedrock): update model YAMLs [bot]

[hganwani-droid]

Auto-generated by poc-agent for provider aws-bedrock.

---

Note

Medium Risk
Primarily updates provider YAML metadata (limits, modalities, params, pricing) which can change routing/validation behavior and displayed costs. Risk is moderate because incorrect values could cause request failures or mis-priced usage for affected models.

Overview
Updates AWS Bedrock model YAMLs to reflect current capabilities, limits, and pricing across several models.

Notable changes include: adding audio I/O and max_tokens param metadata for amazon.nova-2-sonic-v1:0; expanding amazon.nova-pro-v1:0 input modalities (adds video/pdf/doc) and correcting top_k bounds; marking amazon.nova-sonic-v1:0 and amazon.titan-embed-g1-text-02 as deprecated.

Adds/refreshes full definitions (costs, features, limits, params, sources) for newer chat models like anthropic.claude-opus-4-6-v1, minimax.minimax-m2.5, nvidia.nemotron-super-3-120b, and zai.glm-5, and normalizes Stability image model modalities/cost entries (including adding explicit output modalities and updated sources).

^{Written by Cursor Bugbot for commit 160afce. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Missing context_window in Nemotron Super limits

Medium Severity

The newly defined limits section for nvidia.nemotron-super-3-120b includes max_output_tokens and max_tokens but is missing context_window. Every other NVIDIA Nemotron model on aws-bedrock (nemotron-nano-3-30b, nemotron-nano-12b-v2, nemotron-nano-9b-v2) defines context_window in their limits. Per NVIDIA's documentation, this model supports a 1M-token context window, and omitting it means consumers of this YAML won't know the model's total input capacity.

 

[cursor]

Missing max_output_tokens/max_tokens and params for minimax model

Medium Severity

The newly defined minimax.minimax-m2.5 model has thinking: true but its limits section only contains context_window: 204800 — it's missing max_output_tokens and max_tokens, and has no params entry for max_tokens. Every other newly-defined thinking-enabled chat model in this same PR (anthropic.claude-opus-4-6-v1, nvidia.nemotron-super-3-120b, zai.glm-5) includes these fields. Without output token limits, the system cannot properly constrain generation length, which is especially important for thinking models that can produce long reasoning chains.