NexForge: Scaling Agent Capabilities through Requirement-Driven Task Synthesis for LLMs

Benchmark Model Rank Results
computer-use-agents-on-osworld-verifiedNex-N2.5-Pro#4Success Rate: 82.2
deep-research-agents-on-browsecompNex-N2.5-Pro#4Accuracy: 89.7
code-generation-on-swe-bench-proNex-N2.5-Pro#12Resolved: 61.2
code-generation-on-terminal-bench-2-1Nex-N2.5-Pro (NexAU)#15Accuracy: 82.7