Who Is More Honest with JSON: ChatGPT, Gemini or xAI?
Habr developers ran adversarial tests of Structured Outputs from OpenAI, Gemini and xAI to see how faithfully each follows JSON Schema. They asked the models to deliberately break the response format and watched where each provider holds the line. The results were unexpected: the restrictions on anyOf, oneOf, allOf and strict: true differ across all three. The takeaway: for production systems that use several providers, this is critical.
AI-processed from Habr AI; edited by Hamidun News
Habr developers ran a study of how Structured Outputs work at three leading LLM providers: ChatGPT (OpenAI), Gemini and xAI.
How did they test honesty in JSON handling?
The authors used adversarial testing. They built prompts designed to push the model into violating the JSON Schema, then checked where each provider holds the line and where it gives way.
- Nov 1, 2026Local models lose to Opus and Codex: 0.21 on a MacBook versus 1.00 for Codex
- Nov 1, 2026Kimi scores 96.2% on a 1C database analytics test and emerges as a fallback if Claude access is blocked
- Nov 1, 2026OpenAI Rewrites GPT Behavior Rules: Five Values and a Chain of Command
- Nov 1, 2026Google admits AI Mode cuts search revenue, shares fall 15%
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.