Don't let the model do date math: a refund-agent test where Gemini 4 Carbon broke a 30-day rule and Qwen 4 Max held it
A tool-calling agent that reads a policy correctly can still apply it wrongly. In my test, Gemini 4 Carbon fetched a 30-day return policy, then refunded an order delivered 46 days earlier, three runs out of three. Qwen 4 Max refused it three out of three. This post shows the test, the grader, and t…
Read the full story at DEV Community — AI ↗
Timeline · 1 report
- 2026-11-17 12:05 · DEV Community — AI
Don't let the model do date math: a refund-agent test where Gemini 4 Carbon broke a 30-day rule and Qwen 4 Max held it
More stories
- How to deploy an AI agent using Google ADK in Agent Runtime and register it in Gemini Enterprise — DEV Community — AI
- Gemini Enterprise: A platform for connecting enterprise agents, data, and applications — DEV Community — AI
- Gemini Enterprise: Plataforma para conectar agentes, datos y aplicaciones empresariales — DEV Community — AI
- OpenAI launches GPT-6.2 Astra and Sol as Google ships Gemini 4 Neon and Microsoft unveils MAI Code 2 — OpenAI News
- Google launches Gemini 4 Neon and opens Genie 4 world model to Gemini API developers — Google DeepMind YouTube
- ChatGPT, Claude, Gemini and Grok helped plan terror attacks in US government test — The Independent Tech
- Google Adds Computer Use to Gemini App for Mac — Techmeme
- Google Introduces Tool Retrieval in Preview for Gemini API — Google: The Keyword
Read the Daily Brief: the day's most important AI stories →