Vercel

Running Docker containers inside Vercel Sandbox landed in late May, letting agents build and serve containerized apps without touching the host system.¹

Sandbox gained two major capabilities.

• Docker support arrived: install the daemon, build containers, and serve apps from inside the sandbox environment.
• Automatic persistence reached beta: sandboxes now snapshot their filesystem on stop and restore on resume, decoupling storage from compute.² The SDK added writeFiles mode for file permissions and sandbox.update() / listSessions() / listSnapshots() for lifecycle management.

AI Gateway added a broad wave of models.

• MiniMax M3 brings a 1M-token context window with native multimodality and MiniMax Sparse Attention, tuned for agentic web browsing and tool use.³
• GPT-5.5 and GPT-5.5 Pro target long-horizon agentic work across coding and research.⁴
• GPT Image 2 generates up to 2K-resolution images with dense text rendering and multi-style support.⁵
• DeepSeek V4 Pro and Flash offer 1M-token context for coding and high-volume workloads.⁶
• Seedance 2.0 covers text-to-video, image-to-video, and multi-shot generation with synchronized audio.⁷

Vercel Flags reached general availability.⁸ Feature flags with targeting rules, user segments, and environment controls now live in the Dashboard. The Flags SDK integrates natively with Next.js and SvelteKit; an OpenFeature adapter covers other frameworks.

Platform billing and retention updated. Function invocations moved to per-unit pricing ($0.0000006/invocation) for Pro and new Enterprise customers.⁹ Hobby plan deployment retention is capped at 30 days, with the latest preview for open PRs always preserved.¹⁰

Turborepo v2.9.16 shipped cache symlink race fixes, PTY shutdown reliability, Bun workspace dependency version preservation, and heap allocation profiling. The canary series added affected root task detection in query and auth HTTP timeouts.