MaxText adds long-context training support
Google Cloud announced long-context training support in MaxText, its open-source JAX training framework.
Why it matters
The functionality is available in MaxText v0.2.4.
AI news for : 13 updates. Every card links to the maker's own page.
Google Cloud announced long-context training support in MaxText, its open-source JAX training framework.
Why it matters
The functionality is available in MaxText v0.2.4.
Vercel Container Registry now supports compressed image layers of up to 2 GB each, up from 500 MB.
Why it matters
You can manage repositories, images, and tags with the vercel vcr CLI commands in Vercel CLI v58.3.0 or later.
Claude Mythos 5.2 is available to eligible customers in AWS GovCloud (US) Regions through Amazon Bedrock.
Why it matters
Access is limited to allow-listed organizations; to get started, visit Anthropic's Cyber Verification Program.
Token Pass now includes GLM 5.3 Flash, alongside Kimi K3, GLM 5.3, and DeepSeek V4.1 Flash.
Why it matters
Teams can add GLM 5.3 Flash from the Token Pass page in their AI Gateway dashboard.
The Workflow SDK now sends event writes and stream writes to Vercel Workflows over a WebSocket by default, replacing one HTTP request per write, or per write group for streams.
Why it matters
Upgrading to workflow@5.2.0 or later needs no code change, and once you deploy the upgrade, new runs use it.
Claude Fable 5.2 launched, built for demanding reasoning and long-horizon agentic work.
Why it matters
Claude Mythos 5.2 offers the same capabilities only to organizations verified through Anthropic’s verification programs.
Vercel Sandbox now creates snapshots up to 10x faster, with median time to snapshot under 0.4 seconds, down from 4.4 seconds.
Why it matters
Nothing needs to be configured for new snapshots, while existing snapshots keep the previous format.
Google announced the General Availability of Pathways on Google Cloud Platform.
Why it matters
The post explains how Pathways addresses the coordination challenge with one client that describes the whole computation as a graph.
Meta introduced Muse Ember, which it calls its smallest and most efficient model, designed for complex reasoning execution and the work agents repeat all day.
Google AI Studio now lets you build apps powered by Google Maps Platform.
Why it matters
You can describe an app idea and let Google AI Studio handle the configuration and API key management.
Exo agents can now run on Cloudflare, with Durable Objects coordinating conversations and Cloudflare Sandboxes running Linux tools.
Why it matters
You can define an agent in Markdown, then run it with Exo's CLI or HTTP API.
Nanocode coding agents can now run in Cloudflare Durable Objects, with persistent session history and recovery after interruptions.
Why it matters
You can start with the reference Worker and follow Nanocode on Workers for setup, models, and recovery limits.
An experimental Agents SDK example runs the Codex agent loop in a Cloudflare Durable Object.
Why it matters
In this experimental example, conversations, files, and completed model and tool results persist across restarts, allowing the agent to recover after interruptions.
Find dated assistant changes, one table per assistant, in the tracker.