GPT-6 Luna

Tests of Jev and GPT-6 Luna as identity resolution judges showed that a hybrid of Jev and Claude Opus matched frontier accuracy at more than 4x lower cost.

• •Jev and GPT-6 Luna were tested as judges for identity resolution.
• •A hybrid of Jev and Claude Opus achieved frontier-level accuracy.
• •The hybrid configuration operates at a cost more than 4x lower than the alternative.

The performance of Anthropic's lightweight model 'Claude Haiku 5.5' was verified across three Japanese tasks by comparing it with 'Sonnet 5.5' and 'GPT-6 Luna'.

• •The models compared were 'Claude Haiku 5.5', 'Sonnet 5.5', and 'GPT-6 Luna'.
• •It showed performance close to higher-tier models in creating boilerplate documents.
• •Responses fluctuated in tasks requiring answers based on an understanding of one's own stance.

A project benchmarked multiple AI judgment models, including new services from OpenAI and Cloudflare, by having them play Pac-Man.

• •Models such as jev 1.13, kev, clef, GPT-6 Luna, and Laya were tested in Pac-Man ghost battles.
• •These models were confirmed capable of real-time play due to low latency.
• •Each model played 100 games to create a leaderboard, and the repository was open-sourced.