AILobstersabout 3 hours ago
Better Models: Worse Tools
1 min read
Newer Anthropic Claude models, including Opus 4.8 and Sonnet 5, invent extra fields in Pi's edit tool's edits[] array, causing tool call rejections. Older models do not show this behavior. The failure rate reached 20% in some sessions. Stripping thinking blocks halved failures. Strict tool invocation eliminated them entirely.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
