Models that reject the 1M-token context option stop failing every request

Claude Code can ask for a 1-million-token context window (how much conversation the model can hold at once) by sending a beta header, a marker on the request that turns on a trial feature.

If the service rejects a request with that header and the same request works without it, Claude Code now leaves the header out for that model and logs a warning. It sizes the context window without the 1M option. It tries the header again after any of these:

• /clear
• A compaction (summarising the conversation to free space)
• A resume
• A change of provider