Claude Code cloud sessions go live, with a $100 or $250 credit to try them

Claude Code's cloud sessions left research preview on Wednesday, with a one-time credit that's spent before your plan's usage. The same evening, Anthropic said Projects, which coordinates several cloud sessions from one conversation, can now run threads on your own machine, and its engineers said an internal model made claude.ai about 3x faster in two weeks. Its docs now also say Claude Code 2.1.281 reads AGENTS.md, the instructions file other coding agents use, with telemetry off. Three of those claims haven't reached its reference pages yet: the cloud docs skip the credit, the Projects docs still rule out local sessions, and the changelog skips the AGENTS.md fix.

What changed in Claude Code

Claude Code cloud sessions go live, with a $100 or $250 credit to try them

Cloud sessions, which run Claude Code on Anthropic's servers instead of your machine, are "officially available and out of research preview," Anthropic's @ClaudeDevs account posted Wednesday. A session "keeps running after you close your laptop," the docs say, and claude --cloud starts one from the terminal. Existing subscribers get a one-time credit: "$100 on Pro, $250 on Max." Hours later it posted "Sorry for any confusion!" and clarified that the credit isn't extra usage: it's what "your cloud sessions spend first, before falling back onto your normal plan usage." The docs, which don't mention the credit, say cloud sessions "share rate limits with all other Claude and Claude Code usage." Neither post nor the docs say when it expires.

The same evening, @ClaudeDevs said Projects, which runs several cloud sessions (it calls them threads) from one conversation, can now run them "on your own machine." Its docs still say "A local session can't be part of a project." Try it before you plan around it.

Anthropic says Claude made claude.ai 3x faster in two weeks

Anthropic's engineers pointed Claude, running an internal model "roughly comparable to Opus 5.5," at claude.ai and say it got "about 3x faster in a two-week sprint," across "more than three thousand changes" with no rollback. In Claude Code's desktop app, the client's share of a message send fell from 250 to 52 milliseconds for three in four sends. One find: a curly quote or an em dash could freeze the page "for about a second." The prompts, recreated from real threads, are in the post to copy.

Claude Code 2.1.281 reads AGENTS.md with telemetry off, Anthropic's docs say

Claude Code has read AGENTS.md, the instructions file other coding agents use, since 2.1.277 landed on September 18. Sessions through Amazon Bedrock, or with telemetry (usage reporting) turned off, skipped it, and "None of these cases print a warning," a developer found. Anthropic's docs now say that's fixed from 2.1.281. The changelog doesn't say so. The GitHub issue is still open. On an older version, update, or put one line, @AGENTS.md, in a CLAUDE.md, Claude Code's own instructions file.

Claude Code reads AGENTS.md only when telemetry is on

Claude Code 2.1.277 added AGENTS.md support, but the loader sits behind a remote feature flag. With telemetry or nonessential traffic turned off, a local AGENTS.md is skipped without a warning. This is what I measured and the one-line CLAUDE.md I use instead.

blog.szypowi.cz · Hitchhiking through raw bits

Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.

The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.