April 2026 Model Update: GPT Image 2, Grok 4.20 & More
Four new models are now live on TheRouter — including OpenAI's GPT Image 2 (released yesterday) and xAI's Grok 4.20 series with 2M context and multi-agent support.
TheRouter April 2026 model update adds OpenAI GPT Image 2 for reasoning-powered image generation at 2K resolution, xAI Grok 4.20 reasoning/non-reasoning/multi-agent models with 2M context. Model IDs: openai/gpt-image-2, xai/grok-4.20-reasoning, xai/grok-4.20-non-reasoning, xai/grok-4.20-multi-agent.
OpenAI GPT Image 2 — Thinks Before It Draws
OpenAI launched ChatGPT Images 2.0 yesterday (April 21). The API model gpt-image-2 is already routable through TheRouter.
- Native reasoning for images — the model conceptualizes compositions before rendering, producing coherent layouts, accurate text, and precise iconography.
- 2K resolution — up from 1536px. Supports square, portrait, and landscape aspect ratios including 16:9.
- Multi-image consistency — characters and styles stay consistent across a series of generated images.
- Image editing built in — accepts image inputs for transformation, inpainting, and style transfer.
GPT Image 2 Pricing
| Quality | Per Image |
|---|---|
| Low | $0.006 |
| Medium | $0.053 |
| High | $0.211 |
Token-based pricing: text input $5/MTok, image input $8/MTok, image output $30/MTok. Per-image prices above are for 1024×1024; larger sizes cost more.
Quick Start
curl https://api.therouter.ai/v1/images/generations \
-H "Authorization: Bearer $THE_ROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-image-2",
"prompt": "A watercolor painting of a fox reading a book in a library",
"quality": "high",
"size": "1024x1024"
}'xAI Grok 4.20 — 2M Context, Multi-Agent Ready
xAI's Grok 4.20 (released March 31) brings three variants — all with a 2M token context window:
| Model | Best For | Input | Output |
|---|---|---|---|
| xai/grok-4.20-reasoning | Deep analysis, research | $3/MTok | $9/MTok |
| xai/grok-4.20-non-reasoning | General purpose, fast responses | $3/MTok | $9/MTok |
| xai/grok-4.20-multi-agent | Agent orchestration | $3/MTok | $9/MTok |
All three support vision, function calling, structured outputs, and streaming. Aliases like grok-4-20 resolve to the non-reasoning variant by default.
Full Changelog
| Model | Provider | Status |
|---|---|---|
| openai/gpt-image-2 | OpenAI API | New |
| xai/grok-4.20-reasoning | xAI API | New |
| xai/grok-4.20-non-reasoning | xAI API | New |
| xai/grok-4.20-multi-agent | xAI API | New |
| anthropic/claude-opus-4.7 | Anthropic + Bedrock | Already live |
Coming Soon
- DeepSeek V4 — ~1T parameter MoE model with 1M context, expected within weeks. We'll add it on launch day.
- xAI Grok Image & Video — grok-imagine-image ($0.02/image) and grok-imagine-video ($0.05/sec) are on our roadmap.
Getting Started
All new models work with your existing API key — just change the model name.
Questions? Reach out on GitHub.