feat(together-ai): add new models [bot]

[harshiv-26]

Auto-generated by model-addition-agent for provider together-ai.

---

Note

Low Risk
Low risk: adds new Together AI model metadata (pricing/limits/modes) without changing runtime logic. Primary risk is incorrect cost or context-window values impacting billing/quotas downstream.

Overview
Adds four new Together AI model definition YAMLs: Qwen/Qwen3.6-Plus, deepseek-ai/DeepSeek-V4-Pro, mistralai/Voxtral-Mini-3B-2507 (audio transcription), and nvidia/nemotron-3-nano-omni-30b-a3b-reasoning-fp8.

Each file registers the model’s token pricing, optional context window limits, and supportedModes, expanding the provider catalog without modifying application code.

^{Reviewed by Cursor Bugbot for commit a96a8ec. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit a96a8ec. Configure here.}

[cursor]

Audio transcription model uses per-token instead of per-second pricing

High Severity

Voxtral-Mini-3B-2507.yaml is declared as an audio_transcription model but uses input_cost_per_token and output_cost_per_token. Every other audio transcription model in this provider that has costs (openai/whisper-large-v3, nvidia/parakeet-tdt-0.6b-v3) uses input_cost_per_second instead. Using per-token pricing for an audio input model will produce incorrect cost calculations.

 

^{Reviewed by Cursor Bugbot for commit a96a8ec. Configure here.}