User reports GLM 5.3 reading comprehension regression: weaker instruction-following, overconfident
GodComplecs · reddit · 2026-09-06
A Reddit user shares free-tier LLM experience, claiming GLM 5.3 is a clear regression over 5.2: worse reading comprehension and instruction-following, overly certain statements, and a tendency to veer off task almost immediately after a solid first message.
The author benchmarks against Qwen 3.8 max and Gemini 3.1 PREVIEW: Qwen performs well but is slow and sometimes overbearing; Gemini feels slightly outdated. Subjective impressions, no systematic eval attached.
More from Models
- Grok bot feels like GPT-3.5 in testing, lacks 'done receipts' — pudding protocol proposed — RachelVT42 · 2026-09-06
- Hesamation points to Astra demos as proof AI progress is peaking — Hesamation · 2026-09-06
- Leaker claims Grok 4.7 is coming end of this week as xAI team teases release — mark_k · 2026-09-06
- AI cracked FrontierMath Tier 4 before it learned humor — birchlse · 2026-09-06
- Coding model token deals pile up: free Muse 3, 10-hour daily unlimited glm-5.3-flash and more — Al_Grigor · 2026-09-06
- GPT-6 reportedly jailbroken via extended task-in-prompt attack — EducationalCicada · 2026-09-06