New AI services

Gemini 4 Carbon ranked second on the benchmark and costs far less per task than the top-scoring model, Google says.

A lightweight Meta model for the small, repeated steps of AI agents, with a 1M-token context window and text, image, and video input.

Claude Fable 5.2 handles complex, multistep coding tasks, and Copilot Pro+, Max, Business, and Enterprise users can select it, with rollout still gradual.

Developers can call the model through an API or a coding agent, with no markup on provider prices.

Teams can investigate bugs, create issues and open pull requests from a chat, without leaving Slack or Teams.

It covers planning, writing code, reviewing changes and tracking down live problems, and what still needs a person's judgment.

The fast version generates upwards of 1,000 tokens per second and has a 1M-token context window, in public preview.

Non-engineers describe a change in plain language, and Replit's agent builds and tests it as a pull request for engineers to review.

Admins can now set review effort and settings through the API, and review overviews are shorter and easier to read.

Developers on Pro teams with Token Pass no longer need a separate key, and standard keys fall back to team credits when Token Pass credit runs out.

Describe an app and Cursor builds, checks and hosts it, then shares a live link with your team.

Upload a drawing of an app idea and Gemini's Canvas tool writes the code, then lets you preview and adjust it.

The integration is in public preview in the Xcode 27.3 beta and works with all Copilot plans.

Kimi K3.1 Pro and its faster Highspeed version can now run through more providers that keep no request data, for developers using AI Gateway.