DeepSeek Flash v4.1 evals.
v Opus 5.5
20x cheaper
.01 score difference.
First time that it's ready for our swarm lead agent.
Deep dive here: https://evals.agent-swarm.dev/benchmark
evals.agent-swarm.devswarm-evals benchmarkswarm-evals: a frozen, versioned benchmark of coding-agent setups on persistent lead and worker agents.
commit 319d163