Add Gatewayz provider with 58 new models

[vdimarco]

Summary

• Adds the Gatewayz provider along with 58 new model definitions and corresponding provider config. This expands the catalog of gateway-enabled LLMs available to users.

Changes

Provider configuration

• Added providers/gatewayz/provider.toml to register the Gatewayz provider with its API endpoint and environment variable requirements.

Models

• Introduced 58 new model TOML definitions under providers/gatewayz/models, covering a wide range of vendors and model families. Some notable inclusions:
  • Arcee AI: AFM-4.5B
  • Anthropic: Claude Haiku 4.5 and Claude Sonnet 4.5
  • Google: Gemini variants (2.5, lite/preview, flash) and related embeddings
  • OpenAI: GPT-5 family (codex, image-enabled, pro, OSS safeguards, etc.)
  • NVIDIA: Nemotron Nano and related configurations
  • Qwen: Qwen3 VL/Next/Max/etc. in multiple instantiations
  • IBM: Granite 4.0 Micro
  • Meituan LongCat variants
  • MiniMax M2 and M2 Free
  • Perplexity: Sonar Pro Search
  • xAI: Grok 4 Fast
  • zAI: GLM 4.6 and 4.6 exacto
  • DeepSeek, DeepCogito, DeepSeek variants
  • InclusionAI Ling/Ring, OpenGVLab InternVL3, Relace, TheDrummer Cydonia, etc.
• Each model TOML includes metadata (name, release_date, last_updated, attachment, reasoning, temperature, tool_call, open_weights) and limits (context/output) plus modalities (input/output).

Why

• This PR significantly expands the Gatewayz provider’s coverage, enabling users to access a broad spectrum of models through a single provider configuration. It standardizes model entries and keeps the catalog up to date with contemporary models.

Testing plan

• Validate that all new TOML files parse correctly and adhere to the expected schema.
• Load the Gatewayz provider in a local environment and verify that all 58 models appear in the catalog.
• Run sample queries across a mix of modalities (text-to-text, text-and-image to text, image-to-text where applicable) to ensure correct input/output modalities and limits are respected.
• Confirm the provider.toml configuration loads without errors and that API endpoint and env var requirements are correctly wired.

Notes

• There are no breaking changes to existing providers or models.
• This PR focuses on catalog expansion and does not introduce runtime API changes.

Related issues

• No existing issues targeted by this change; intended to enrich the model catalog and provider capabilities.

🌿 Generated by Terry

---

ℹ️ Tag @terragon-labs to ask questions and address PR feedback

📎 Task: https://www.terragonlabs.com/task/510b58a7-234f-43d5-88d8-64f12fd09ecd

---

Note

Registers the Gatewayz provider and adds a large set of vendor model TOML definitions (text and multimodal) with metadata, limits, and modalities.

• Provider:
  • Register providers/gatewayz/provider.toml with API base (https://api.gatewayz.ai/v1), env var GATEWAYZ_API_KEY, docs link, and npm client reference.
• Models:
  • Add numerous TOML model definitions under providers/gatewayz/models with name, release_date, last_updated, attachment, reasoning, temperature, tool_call, open_weights, limit.context/output, and modalities.
  • Include both text-only and vision-capable variants; some with "thinking"/reasoning and "free"/"exacto" profiles.
  • Key vendors/families: OpenAI (GPT-5, o3/o4 Deep Research, embeddings), Google (Gemini 2.5 variants), Anthropic (Claude Haiku/Sonnet 4.5), Qwen (Qwen3 VL/Next/Plus/Max, coder), NVIDIA (Nemotron/Nemotron Nano), Amazon (Nova Premier), Perplexity (Sonar Pro Search), Baidu (ERNIE), MiniMax, Meituan (LongCat), IBM (Granite 4.0 Micro), OpenGVLab (InternVL3), DeepSeek, Deep Cogito, Arcee AI, LiquidAI, Relace, TheDrummer, Z.AI.

^{Written by Cursor Bugbot for commit ed3ab4a. This will update automatically on new commits. Configure here.}

[cursor]

Bug: Context limit mismatch disrupts model prompts workflow

limits.context is set to 0, which effectively disables usable context for the model. This is inconsistent with similar Qwen VL models (e.g., 131072) and will likely cause incorrect behavior or rejection of prompts.