Introducing GPT-6 Sol and Luna

Benchmark Model Rank Results
agents-on-automationbench-private-v1-0-6GPT-6 Luna (max effort) (Zapier AutomationBench)–Success Rate: 20.7
agents-on-automationbench-private-v1-0-6GPT-6 Sol (xhigh effort) (Zapier AutomationBench)–Success Rate: 33.18
code-generation-on-deepsweGPT-6 Luna (max effort) (mini-swe-agent)–Pass@1: 0.6659
code-generation-on-deepsweGPT-6 Sol (max effort) (mini-swe-agent)–Pass@1: 0.6881
computer-use-agents-on-osworld-2-0-0808-offline-82-tasksGPT-6 Luna–Partial: 52.68
computer-use-agents-on-osworld-2-0-0808-offline-82-tasksGPT-6 Sol–Partial: 64.43