Vercel AI Gateway adds Inkling-Small multimodal model

// 1h agoPRODUCT UPDATE

Vercel AI Gateway adds Inkling-Small multimodal model

Vercel announced that Inkling-Small, an open-weights multimodal AI model from Thinking Machines, is now available on Vercel AI Gateway using the model string thinkingmachines/inkling-small. Operating at roughly one-fourth the size of the flagship Inkling model, Inkling-Small provides strong reasoning across text, image, and audio inputs with lower latency and compute costs.

// ANALYSIS

Bringing Inkling-Small to Vercel AI Gateway provides developers with a powerful, budget-friendly open-weights option for production multimodal applications.

* **Resource Efficient**: Delivers performance matching the larger Inkling flagship model at approximately one-quarter of the footprint, drastically cutting latency and token costs.

* **Seamless Integration**: Developers can call `thinkingmachines/inkling-small` instantly within Vercel AI Gateway without needing to manage dedicated server infrastructure.

* **Versatile Multimodality**: Supports text, image, and audio processing natively to streamline building diverse AI features.

// TAGS

vercelvercel-ai-gatewayinkling-smallthinking-machinesopen-weightsmultimodalllm

DISCOVERED

1h ago

2026-07-30

PUBLISHED

2h ago

2026-07-30

RELEVANCE

8/ 10

AUTHOR

vercel_dev

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS14m ago

OWASP Details LLM Prompt Injection Defenses

The OWASP LLM Prompt Injection Prevention Cheat Sheet delivers actionable security guidance to defend Large Language Model applications and autonomous AI coding agents against direct and indirect prompt injection attacks. It outlines a multi-layered defense model featuring strict input validation, least-privilege tool access, structural prompt segregation, and deterministic guardrails.

UPDATE30m ago

Vercel Sandbox adds multi-agent user isolation

Vercel has introduced support for running multiple isolated AI agents inside a single Vercel Sandbox, with each agent operating as a distinct Linux user. This update enables developers to execute complex multi-agent workflows—such as coordinated coding, planning, and code review agents—within shared sandbox infrastructure while maintaining strict OS-level privilege separation and security boundaries.

MODEL46m ago

NVIDIA Cosmos3-Super-4Step tops open-weights leaderboards

NVIDIA has released Cosmos3-Super-Text2Image-4Step and Cosmos3-Super-Image2Video-4Step, distilled 4-step variants of its 64B Cosmos3-Super foundation model family. Evaluated on the Artificial Analysis Arena, these distilled models achieve state-of-the-art results among open-weights releases, securing the top ranking for Image-to-Video and third place for Text-to-Image while significantly accelerating generation speeds.