Community · Oct 31, 2026, 16:33 UTC

Gear, an open-source RSI framework, reports GPT 5.6 Luna gains on agent benchmarks

The project reports Luna with an optimized harness rising from 27.0% to 40.0% on AutomationBench Marketing and 52.8% to 62.9% on Terminal-Bench 2.1, with results from its own evaluation. At max effort, Luna's estimated API cost averaged $0.0568 per task versus $1.1858 for Astra, excluding optimizat…

• ModelGPT 5.6 Luna
• ModelAstra
• ModelGEPA
• Topicagents
• Topicbenchmark
• Topicresearch
• Topicopen-weights