Gemini Frequently Hallucinates During Code Comprehension
GlompSpark · reddit · 2026-07-17
While testing the coding analysis capabilities of Gemini Flash 3.5, a user found it continuously hallucinating while reading a 7k-line code file:
- First, it falsely claimed that a certain logic was inside the realm class
- Then it fabricated non-existent header files and functions
- Even after being reminded, it continued to generate plausible but incorrect explanations
- Finally, it even admitted it hadn't actually seen the relevant code and was just "pretending to understand"
The key point of the post is that this version of Gemini is highly unstable on tasks involving long code files and locating specific logic, resulting in an experience of continuous hallucination rather than a one-off mistake.
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26