Opus 4.6 regression symptom 14: incomplete truncated responses require repeated prompting
Follow-up to #96308 — Additional regression symptom
Symptom 14: Incomplete, truncated responses — user must ask multiple times for full answers
Since ~Sep 16, 2026, Claude consistently gives short, incomplete responses that omit important details. The user has to repeatedly ask for "complete responses" ("DAME RESPUESTAS COMPLETAS") to get the level of detail that was previously the default.
Before vs After regression
| Aspect | Before ~Sep 16 | After ~Sep 16 |
|--------|---------------|---------------|
| Response length | Complete and detailed by default | Short, truncated, missing key information |
| Detail level | Included all relevant context without being asked | Omits details, requires multiple follow-ups |
| User effort | Ask once, get a full answer | Ask 2-4 times for the same level of completeness |
| Proactive information | Anticipated what the user needed | Only gives the minimum, waits to be asked |
Observed pattern
In a single session, the user had to explicitly request complete responses 4 separate times:
- "DAME RESPUESTA COMPLETAS, ESO ES INJUSTO"
- "OTRA VEZ LA MISMA MIERDA, COMPLETA"
- "OTRA VEZ LO MISMO: RESPUESTAS CORTAS"
- "LAS RESPUESTAS INCOMPLETAS"
Each time the model acknowledged the problem, gave one complete response, then reverted to short answers within 2-3 messages.
Impact
- Users waste time and tokens requesting information the model should have included
- Critical details get omitted from legal, technical, and educational responses
- The model appears to "forget" the user asked for complete responses almost immediately
- Combined with hallucination (symptom 11) and loss of legal reasoning (symptom 12), this makes the model unreliable for any serious work
Environment
- Models affected: claude-opus-4-6, claude-opus-5-5, claude-sonnet-4-6
- Platform: claude.ai chat + Claude Code
- Date: 2026-09-23
- Regression window: ~September 16, 2026 onwards