Unverified50% confidenceFactExact time
The tests designed three real-world scenario tasks: Kanban application development, Blender lantern scene modeling, and a Godot 3D platformer game, with the prompts for Blender and Godot not publicly disclosed and not having been trained on.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the Godot game test, Q3 repeatedly caused the Godot program to crash and could not complete the task, Q2 fell into a loop and failed, and Q1 entered a loop immediately after connecting the tools.63% similarUnverifiedGODMODE is essentially an application-layer jailbreak tool that does not touch the model itself, but instead induces behavioral deviation in the model at the application layer through carefully crafted inputs.61% similarUnverifiedThe test scenario was a game project combining elements of Minecraft and No Man's Sky, with bugs involving items in the inventory system that could not be used and could not be moved to the hotbar.57% similarUnverifiedIn the dungeon exploration game raycasting programming task, both models required multiple rounds of communication and ultimately came out roughly even.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/907323API
curl https://kongchang.com/api/v1/knowledge/claims/907323MCP
get_claim(id=907323)