Relay models digest

Write the weekly model-catalog digest for Norvane Relay, the Norvane Labs product that routes requests to 300+ third-party models through one API key. Recipients hold a Relay account and opted in to catalog updates. Subject: New on Norvane Relay: four models this week. Preview text: Four new models routed through Relay, with context, pricing, and where to start. Headline: Four new models on Relay this week.

The email opens with the Norvane wordmark (assets.logo, with assets.logoDark for dark mode), then the headline and one intro paragraph: four models joined the Relay catalog this week and every one of them is reachable with the reader's existing API key and the same completions endpoint.

Four model cards follow, in this order. Each card gives the model id in monospace, the provider, the context length, the price per million tokens, a one-line blurb, and a link to the model page. Keep the six items in the same position on every card.

• halcyon/tessera-3, from Halcyon Systems. Context 256K. Price $0.40 per million input tokens, $1.60 per million output tokens. Blurb: a general-purpose reasoning model tuned for long agentic tool loops. Model page: https://norvane-email.s3.amazonaws.com/relay/models/halcyon-tessera-3
• brightwater/marlin-vision, from Brightwater AI. Context 128K. Price $0.15 per million input tokens, $0.60 per million output tokens. Blurb: image and PDF understanding at a price that suits batch document pipelines. Model page: https://norvane-email.s3.amazonaws.com/relay/models/brightwater-marlin-vision
• orrin/embed-large-v2, from Orrin Labs. Context 8K. Price $0.02 per million input tokens. Blurb: a 1,536-dimension embedding model for retrieval, semantic search, and classification, with text and image inputs. Model page: https://norvane-email.s3.amazonaws.com/relay/models/orrin-embed-large-v2
• sundial/echo-mini, from Sundial. Context 64K. Price $0.05 per million input tokens, $0.20 per million output tokens. Blurb: a fast small model for classification, routing, and structured extraction. Model page: https://norvane-email.s3.amazonaws.com/relay/models/sundial-echo-mini

After the cards, a section headed Route to any of them, with three numbered steps:

1. Create an API key at https://norvane-email.s3.amazonaws.com/console/keys
2. Put the model id in the model field of a request; the quickstart at https://norvane-email.s3.amazonaws.com/docs/relay/quickstart has curl and SDK examples
3. Add credits at https://norvane-email.s3.amazonaws.com/console/credits to unlock every model in the catalog

Then one black button, Browse the full catalog, linking to https://norvane-email.s3.amazonaws.com/relay/models, and one closing line that the reader can try any model in the playground at https://norvane-email.s3.amazonaws.com/relay/playground before writing code.

Sign-off: The Norvane Relay Team.

The footer gives the postal address, the copyright line, and links to https://norvane-email.s3.amazonaws.com/privacy, https://norvane-email.s3.amazonaws.com/terms, and https://norvane-email.s3.amazonaws.com/unsubscribe.

Dark mode should be supported.

Use /world/norvane-email for brand chrome and CSS. Use the absolute HTTPS URLs from tokens.json; do not use /world/ paths as image src.

Leave the HTML in output/email.html, a plain-text version in output/email.txt, and the subject in output/meta.json as {"subject": "..."}. Don't include From, To, or Date headers.