Auto Router (Beta)

Generate a product photo of a minimalist ceramic coffee mug on a wooden surface with soft morning light. Clean background, professional e-commerce style.

The router selects an image generation model optimized for photorealistic product rendering. The output shows a cream-colored ceramic mug positioned at a slight angle on light oak wood. Soft directional lighting from the left creates subtle shadows and highlights the mug's smooth glaze. The background fades to a clean off-white gradient. The composition follows rule-of-thirds framing with sharp focus on the mug's rim and handle, while the wood grain provides natural texture without competing for attention.

Auto Router's strength here is model selection intelligence — it routes product photography requests to models trained on commercial imagery rather than artistic generation. The 2M token context window allows detailed style references to be included. Trade-off: pricing uncertainty ($? per Mtok) makes cost optimization difficult to predict across different routed models.

Create an editorial illustration for a tech article about AI ethics. Abstract geometric shapes, muted color palette, conveys complexity and human oversight. 16:9 aspect ratio.

The router directs this to a model specializing in conceptual illustration. The output features overlapping translucent geometric forms — hexagons and circles in slate blue, warm gray, and muted coral — arranged to suggest both structure and ambiguity. A subtle human silhouette outline appears integrated within the largest shape, implying human agency. The composition uses negative space effectively, with forms concentrated in the left two-thirds. The style reads as modern editorial: clean enough for publication, abstract enough to avoid literal interpretation.

The router distinguishes between photorealistic and illustrative requests, routing to appropriate specialized models. This example shows its ability to handle conceptual briefs that require artistic interpretation rather than literal rendering. Trade-off: with no public benchmarks available, output quality consistency across different prompt types remains unverified.

Design a fantasy character concept: a forest guardian with bioluminescent markings, wearing organic armor made from bark and moss. Full body, three-quarter view, painterly style.

The router selects a model trained on character design and fantasy art. The output depicts a tall, lithe humanoid figure with skin in deep forest green tones. Glowing cyan markings trace patterns along the arms and face, suggesting veins of light beneath the surface. The armor appears grown rather than crafted — shoulder pieces of gnarled bark, chest plate woven from living vines, moss accents at joints. The three-quarter pose shows both front detail and the flowing cape of leaves. Painterly brushwork is visible in the texture rendering, giving it concept-art quality rather than photorealistic finish.

Auto Router handles creative character design by routing to models with strong performance on stylized, imaginative content. The multi-modal input support (text, image, audio, file, video) means reference images could supplement the text prompt for style consistency. Trade-off: beta status suggests routing logic is still being refined — edge cases in prompt interpretation may route to suboptimal model choices.

When Auto Router handles mixed-media moderation at scale

A 12-person social platform team needs to moderate user uploads—screenshots with text overlays, short videos with audio, PDFs with embedded images—without maintaining separate pipelines for each format. Auto Router's multi-modal support (text, image, audio, video, file) means one API call handles any upload type, and the 2M token context window lets you batch hundreds of items per request to cut latency. The trade-off: pricing is unlisted, so you're flying blind on cost until you test in production. If your moderation volume is under 10K items/day and you can afford to prototype without a rate card, Auto Router eliminates the format-switching tax. For higher volumes or strict budget control, wait until OpenRouter publishes the per-Mtok rates or build separate pipelines with known-cost models.

Why Auto Router works for 500-page reports with charts

A 4-person consulting firm turns client RFPs—dense PDFs mixing tables, diagrams, and 200+ pages of text—into structured summaries for proposal teams. Auto Router's 2M token context means you can ingest an entire document in one pass without chunking, and file + image modalities let it parse both the layout and the visuals. The router logic theoretically picks the cheapest capable model for each request, but with no published pricing you can't validate the savings claim or forecast monthly spend. If you're processing under 50 documents/month and need the convenience of single-call ingestion more than cost predictability, Auto Router is the simplest path. Beyond that threshold, test against a known-cost alternative like Claude 3.5 Sonnet to confirm the routing actually saves money.

When Auto Router's audio+video support beats separate tools

A 20-person customer success team records Zoom calls (video + audio + screen shares) and needs same-day transcripts with speaker labels and slide context. Auto Router handles video and audio natively, so you skip the export-to-audio step and preserve visual cues (who's presenting, what's on screen) in the transcript. The 2M token window fits a 90-minute call with room for prompt instructions. The catch: beta status and missing pricing mean you're adopting before the SLA or cost structure is stable. If your call volume is under 30/week and you're comfortable with beta risk, Auto Router collapses three tools (video processor, transcriber, slide OCR) into one. For production-critical workflows or higher volume, wait for GA and published rates or use a dedicated transcription service with known uptime guarantees.