Yeah, it's an interesting test because he gave Kimi credit for running inside the harness he liked, which I presume is Claude Code. But he also said it was noted for its harness sensitivity, and then said it didn't do as well as Opus.
A lot of folks don't seem to understand that it's the harness and its behavior paired with a particular model that produces the result, and you have to focus on both aspects. He barely acknowledges the importance of the harness in his write-up, beyond the two notes above.
There's a very interesting benchmark comparison below of Opus 4.7 run under three different harnesses : OpenCode, Cursor and Claude Code where it's not very close at all and Opus's native harness, Claude Code, performs worst of all three.
The pass@1 scores are 50/45/40 for OpenCode/Cursor/Claude Code respectively.
Comments
Yeah, it's an interesting test because he gave Kimi credit for running inside the harness he liked, which I presume is Claude Code. But he also said it was noted for its harness sensitivity, and then said it didn't do as well as Opus.
A lot of folks don't seem to understand that it's the harness and its behavior paired with a particular model that produces the result, and you have to focus on both aspects. He barely acknowledges the importance of the harness in his write-up, beyond the two notes above.
There's a very interesting benchmark comparison below of Opus 4.7 run under three different harnesses : OpenCode, Cursor and Claude Code where it's not very close at all and Opus's native harness, Claude Code, performs worst of all three.
The pass@1 scores are 50/45/40 for OpenCode/Cursor/Claude Code respectively.
https://artificialanalysis.ai/agents/coding-agents#harness-c...
I've seen other benchmarks where Pi also outperforms Claude Code both in model performance and in much reduced token usage.