Vercel

Vercel is running GPT-5.6 Sol at half price on its AI Gateway through September 18, across every service tier including fast mode, and on cached tokens, cache writes and long context requests. Gateway billed traffic only, and the model ID is unchanged.

Vercel's open-source fx coding agent now runs behind the AI SDK harness layer via the @ai-sdk/harness-fx adapter, connecting over the Agent Client Protocol. It joins Claude Code, Codex, Cursor, Cline, OpenCode, Grok Build, Deep Agents, and Pi under the one HarnessAgent API.

OpenAI cut list pricing for GPT-5.6 Sol, and the standing gateway discount now applies on top of the lower rate through September 18: input down a fifth, output down a third, on every service tier. The model ID is unchanged, so existing traffic reprices on its own.

Google's speech-to-text model lands on Vercel's AI Gateway in batch and WebSocket variants, with AI SDK V7 adding streamTranscribe for live microphone input. It detects language on its own across 85+ languages, follows mid-sentence switches, and takes custom vocabulary for names and jargon.

Vercel's AI Gateway can now route any coding agent through one command, centralising keys, budgets, traces and spend, and letting an agent use any of 200+ models including ones it has no native support for.

Vercel's October AI Gateway index reports open-weight models at 71.5% of tokens and 23% of spend, spend up about 13x since December, and nearly a third of teams switching primary lab in September. Data reflects Vercel's customer base only.

Vercel Sandbox creation now waits for snapshot restoration, so a returned sandbox is immediately ready and startup errors reject at creation. Creation is about 50 ms slower at p50, with no SDK change needed.

The Slack adapter for Chat SDK now handles org-wide Enterprise Grid installs: installations keyed by enterprise ID, per-installation profile and mention caches, automatic team_id on workspace-scoped calls, 24-hour redelivery dedup, and W-prefixed Grid user IDs. Single-workspace installs unchanged.

A postmortem tracing an RCE found in the AVIF decoder libheif through Next.js's image pipeline (sharp, libvips), covering reproduction, cross-maintainer disclosure, and the platform-wide mitigation deployed before an upstream fix landed.

The skills CLI can now install agent skills directly from Notion pages via a new Agent Skills API, letting teams author and update reusable skills in Notion and push them into any compatible agent without a Git repository.

Vercel Sandbox observability now reports memory usage (average, P75, P95) alongside CPU and data transfer, in the dashboard, CLI, and query builder, so teams can see when a sandboxed agent run is approaching its memory ceiling.

AI SDK 6.0.313 processes only the first approval response per tool call, removes orphaned reasoning when pruning tool calls, and keeps repaired tool names in streamed UI message parts.

GLM 5.2, Z.ai's open-weights coding model with a 1M-token context window, is free for eve agents through 27 August via Blackbox on Vercel's AI Gateway, and is the default model for new agents.

Vercel's Token Pass subscription now includes DeepSeek V4.1 Flash alongside Kimi K3 and GLM 5.3. The $30/month Pro-team plan grants $80 in credit through AI Gateway and works with agents such as Pi, Cline and fx.

Vercel Connect adds a TanStack AI integration letting agents call OAuth-protected MCP servers, such as Linear, with token refresh and per-user consent handled automatically rather than managed by hand.

DeepSeek V4 Pro on Vercel's AI Gateway now serves updated weights by default under the unpinned model ID, so existing calls pick them up silently; pin the dated ID to fix the version. The changelog's own code sample is inconsistent about that suffix.

Vercel adds Amazon S3 to its Marketplace and v0. Resources are created from the dashboard, connection details are injected as environment variables, and access uses short-lived OIDC credentials rather than stored AWS keys.

Vercel's AI Gateway now serves Claude Fable 5.1, and the changelog documents the fallback pattern that matters for it: list backup models in providerOptions.gateway and the gateway walks the list when a safety classifier refuses. Works across Chat Completions, Messages, and Responses.

Vercel's AI Gateway adds an HTTP API and native TypeSafe client support for the Jev decision model, alongside its existing AI SDK path, unifying billing and usage tracking across all three integration routes.

AI Gateway requests now land on a dedicated logs page showing cost, token counts split across input, output, reasoning and cache, duration, and which model, provider and region served each call. Open a request and you get the complete fallback path, including why each failed attempt failed.

Meta Superintelligence Labs' first image model is now callable through AI Gateway as meta/muse-image-1.0. One model covers both text-to-image and instruction-based editing — pass reference images in prompt.images to steer style, or hand it a file with an edit instruction and it changes only what you

Qwen 3.8 Flash is live on Vercel's AI Gateway: text and image input, a 1M-token context, up to 65k output tokens, and Alibaba's recommendation to use it for coding, tool calls and multi-step agent workflows. BYOK requests carry no platform fee.

One command now takes a repo from unconfigured to a finished security review. Vercel's deepsec init creates an isolated workspace, configures model access, maps the codebase's attack surface, generates extra scan patterns where coverage is thin, and checkpoints each step so interrupted runs resume.

Vercel argues machine-to-machine auth still resembles pre-HTTPS web, with static keys the default. It introduces OIDC UI, a shadcn registry for trust-policy forms and token verification, widens OIDC across its platform, and sets new auth requirements for Marketplace providers.