OpenAIAnthropicGooglexAIPerplexity
← Back to all updatesPublished ·Anthropic• Minor
Anthropic launched mid-conversation system messages for Claude Opus 4.8, without a beta header.
The read
Instruction changes during long-running sessions no longer have to break prompt cache hits.
The frame
For builders, instruction changes now sit after a user turn in messages, subject to placement rules.
Primary source
Anthropic →Keep reading
See what Nova3 builds →Related updates
This is the world we build in. It moves this fast. We move with it.
Bring us your project →