AI intelligence

Vercel has added OpenAI’s GPT-6.1 Sol to AI Gateway through the AI SDK and OpenAI-compatible APIs. The announcement positions it for coding, computer interaction and document-heavy professional workflows, including tasks involving PDFs with tables and charts. Vercel also describes lower standard input and output pricing than GPT-6 Astra. These capability comparisons are publisher claims, rather than independent evaluation results.

AI Gateway User Insights now gives you more context about the traffic flowing through your gateway. It shows what users and agents are doing with AI, and where a selected model may be more capable than a task requires. On the analysis side, User Insights groups conversations by task, tracks conversation turns, and helps you compare model fit with cost and latency. The Potential Savings view highlights requests that may work with faster or less expensive models without compromising output quality. These are the same signals that Cloudflare's Auto Router uses to select a model based on task and cost. These new insights are available to all AI Gateway customers at no additional cost. For more information, refer to User Insights.

Sep 28, 22:08 UTC Resolved - On September 28, 2026, between 19:23 UTC and 22:08 UTC the Copilot Code Review service was degraded and pull request reviews did not complete in all environments.

Google announced that filmmaker Jeff Synthesized won the Future Vision XPRIZE with The Gifted, a fictional story about a child recreating his late mother through technology. The project receives a $100,000 prize and $2.5 million in feature-production funding. Google’s 100 ZEROS initiative and Range Media Partners will help develop the film, which was selected from more than 2,500 entries.

Google recommends combining frontier-model APIs with smaller open-weight models instead of using one model for every production task. Its article presents Gemma for local execution, repetitive routing and task-specific tuning, with larger models reserved for demanding synthesis. Customer examples describe latency and cost improvements, but these are attributed case studies rather than evidence that the same architecture will benefit every workload.

Mistral opened a Munich hub for industrial and physics-focused AI research, with applied engineers supporting enterprise partners. The company says it is working with BMW on crash simulation and Siemens Energy on industrial applications, alongside a Technical University of Munich research partnership on aerodynamic digital twins. These efforts extend its industrial research program; the announcement does not establish independently measured production gains.

Vercel added Anthropic’s Claude Sonnet 5.5 to AI Gateway for coding, document creation and other scoped tasks. It is accessible through the AI SDK and OpenAI- or Anthropic-compatible APIs, and supports Zero Data Retention on the gateway. Vercel describes better multi-step coding, desktop interaction and document output than Sonnet 5.

Fireworks’ Ember-1 is available through Vercel AI Gateway as a research preview with an initial two-week window. Built on Kimi K3, it supports a one-million-token context, image input and tool calling. Fireworks reports roughly 40% fewer generated tokens at comparable quality in its evaluations. The endpoint supports Zero Data Retention and No Prompt Training.

Vercel’s analysis of skills.sh describes rapid growth in reusable instructions for AI agents, with software engineering leading published listings but demand spread across more kinds of work. Install activity is concentrated among a small share of skills. The report uses aggregate registry counters, which do not represent unique users, and its category findings come from a classified sample rather than the whole catalog.

Vercel introduced Pixel Canary on AI Gateway as a stealth model with temporarily free access. The announcement emphasizes coding and interface development, citing Next.js evaluation results that allow up to four attempts per task. Zero Data Retention is unavailable, and submitted prompts and responses may be used for training and model improvement.

Google has added Live Avatar to Gemini 3.8 Live in Gemini Enterprise, pairing conversational audio with generated video. The feature can continue dialogue while tools run in the background and, according to Google, supports synchronized speech and expressions across 97 languages. Preset avatars are available, while custom avatar creation requires enterprise allowlisting; generated audio and video carry SynthID watermarks.

Vercel Connect now supports TanStack AI agents calling OAuth-protected MCP servers. A new transport adapter obtains authentication through Connect before each request, handling token freshness without applications storing or rotating those credentials. If a user has not granted access, client creation raises a consent challenge before the model runs, allowing the application to redirect the user to the authorization flow.

Google is expanding Beam video-conferencing hardware to six countries through HP and its channel partners, with support for Google Meet and Zoom. A partnership with Industrious adds bookable installations at selected U.S. workspaces starting in October. Google also reports better connection and fewer follow-up meetings in an eight-week internal study; those findings are company research, not a guarantee for every workplace.

Google outlined a planned persistent-memory layer for Private AI Compute to retain assistant context across devices. Its design combines encrypted cloud storage, keys held on users’ devices and isolated computing environments that temporarily process the data. The company also published updated technical material and server-verification information, positioning the architecture as a way to add continuity without giving Google access to stored personal context.

Google introduced Gemini 3.8 Flash TTS for detailed voice direction and Flash-Lite TTS for higher-volume speech generation. Both are rolling out through the Gemini API and AI Studio, while enterprise API access is described as coming later. Voice replication requires a matching consent recording and has regional restrictions; Google says generated audio carries SynthID watermarks. Voice remixing remains a forthcoming feature.

Vercel added Google’s Gemini 3.8 Flash TTS and Flash-Lite TTS to AI Gateway. Both generate speech in more than 100 languages and support long narration and dialogue between two speakers. Vercel positions Flash-Lite for high-volume generation and Flash for more expressive character voices. The documented AI SDK integration requires version 7 or later.

Claude Opus 5.5 is available through Vercel AI Gateway with a one-million-token context and up to 128,000 output tokens. Integrators must account for two compatibility changes: thinking is always adaptive, and forced tool selection is unsupported. Requests that disable thinking, specify a fixed reasoning budget or require a particular tool can fail. Vercel recommends structured outputs for workflows that previously forced tools to obtain JSON.

Vercel added OpenAI’s GPT-6 Sol and GPT-6 Luna to AI Gateway. The announcement positions Sol for extended coding investigations and more complex agent workflows, while Luna targets focused, repeatable tasks at higher volume. Vercel describes both as lower-priced than GPT-6 Astra and claims improved reliability over their GPT-5.6 counterparts.

Vercel AI Gateway now exposes TypeSafe AI’s Jev through existing TypeSafe clients and an HTTP API, alongside the AI SDK. Jev returns typed decision results rather than generated prose, supporting boolean, choice and scoring questions. Existing clients can retain their evaluation calls while changing connection settings, and requests through all three interfaces appear in the gateway’s billing and observability.

Vercel added Xiaomi’s MiMo V2.6 Pro, Flash and Pro UltraSpeed to AI Gateway. All three accept text, images, audio and video, with a one-million-token context window and up to 128,000 output tokens. Pro targets complex work, while Flash is positioned as a lower-cost option. Vercel says UltraSpeed serves the same Pro model with up to twentyfold faster output.

Vercel added Grok 4.7 to AI Gateway for extended coding, document and presentation work. The model offers a 500,000-token context window and four reasoning levels, letting applications choose how much reasoning to request. It can also be selected in Vercel’s eve agent framework. The launch promotion offered 40% off only through September 27, so that advertised discount has expired.

Google is expanding its AI & Economy Research Program with additional academic advisers and research leaders. Anu Madgavkar and Daniel Rock will help direct work on adoption, labor markets, enterprise productivity and scientific discovery, while Philippe Aghion and Ajay Agrawal join advisory and visiting roles. The announcement describes a research agenda intended to inform future ATLAS updates, rather than new economic findings.

Google describes two fashion-design collaborations using custom Flow tools ahead of New York Fashion Week. Jane Wade’s styling tool let her combine garments and accessories on digital models before producing additional samples, while Sergio Hudson used simulated runway layouts, lighting and props to plan within his budget. These are specific designer workflows; the article does not quantify general productivity or cost savings.

Vercel reports that TypeSafe AI’s Jev reached nearly 13% of AI Gateway’s paid teams within its first day, more than twice the adoption of any previous model launch on the service. Jev is designed for software decisions, returning typed answers and probabilities instead of prose. The report describes early uptake and leaves the durability of that adoption as an open question.