ifm k2 horizon 7b swe bench
deepseek v4 flash agent benchmarks
prime intellect unifies 365000 agentic tasks
swe bench verified
senior swe bench
statewright visual state machines
tomasullm out of order speculative execution for llm agents
modularrsi modular recursive harness self improvement
swe prime fewer trajectories better performance
swe bench science
swe pruner pro