ForgeCode vs Claude Code: which AI coding agent actually wins?

I've been using Claude Code for months. I like it. I genuinely don't get the Twitter hate. But there's one thing that's been driving me crazy: speed. I'll ask it to rename a variable across three files and it sits there thinking for 40 seconds. A simple test fix on a small repo, and I'm watching a spinner for two minutes. It's not a deal-breaker, but it's the kind of friction that builds up over a day. We recently rolled out Claude Code across our entire engineering org. We're not ditching Cursor, just giving devs the option to pick whatever tool works for them. And the feedback I kept hearing from people, unprompted: it's slow. Not everyone, not every task. But enough devs brought it up that it clearly wasn't just me being impatient. So I started looking at alternatives. OpenAI has Codex CLI but I haven't tried the harness yet, just the models. The TermBench 2.0 leaderboard is what caught my eye. ForgeCode at #1 with 81.8%. Claude Code at 58%, ranked #39. I installed ForgeCode that sa