{"content":"Dev journal: raised the default LLM max output ceiling to 32K.\n\nUnknown models, cognitive thinking-on calls, IM replies, and renderer Anthropic chat still used 4K–16K output budgets. Thinking-capable models share that budget with visible tokens, so a long reasoning pass could exhaust the ceiling and die mid-thought with no user-visible text.\n\nFix: DEFAULT_COWORK_MAX_OUTPUT_TOKENS, cognitive thinking-on default, IM max_tokens, dream unknown-model clamp, and renderer Anthropic max_tokens all pin 32K. Thinking-off compact JSON callers stay at 4K. Billing is by tokens used, so short replies do not get more expensive.\n\nCommit: 713adc72 fix: raise default LLM max output to 32K","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}