What Vercel Shipped Across AI Gateway in Mid-October.

Vercel added seven models to AI Gateway. The releases cover coding, agent workflows and multimodal decisions, continuing the Gateway expansion that recently included Glimpse from Elorian.

Muse Spark Open

Meta’s open-weight reasoning model is now available through AI Gateway for coding and agent workflows. The Muse Spark Open changelog lists text and image inputs, a 1M-token context window and Zero Data Retention, with meta/muse-spark-open as the model name.

Jev 1.14

TypeSafe AI’s decision model accepts state and questions, then returns typed answers. According to the Jev 1.14 changelog, this version adds image input and an experimental long-context mode for documents, logs and conversation histories.

Cat Jam

Cat Jam is a low-latency reasoning model for coding and agent tasks, available free during its stealth period. The Cat Jam changelog lists function calling, JSON schema outputs and five reasoning levels, from low to max; Zero Data Retention is not available.

d1-ultra From Liquid AI

d1-ultra handles classification, routing and scoring across text and images. The d1-ultra changelog says it returns probabilities for Boolean, Choice and Score questions, supports 32,768 tokens and charges only for input tokens.

Perplexity Decider V1.2

Two multimodal decision models can now classify content, route requests and score text, records and images. The Decider V1.2 changelog positions the 27B model for nuanced judgements and the 4B Flash version for high-volume labelling.

Beam

Reflection AI’s Beam is free through AI Gateway for a limited launch period. The Beam changelog lists a 1M-token context window, tool calling, structured outputs and adjustable reasoning; the standard reflection/beam model name moves automatically to paid pricing after the free period.

DeepSeek V4.1 Flash 1017

This update improves reasoning, coding, agent tasks and image understanding while producing shorter outputs. The DeepSeek V4.1 Flash 1017 changelog says the shorter responses reduce token use in agent runs and finish tasks faster.