tbench benchmark?

[Kreijstal]

What feature would you like to see?

How good is kimi k2 on tbench?

Additional information

with terminus got 35% but gpt5 as well, I am using kimi k2 thinkng with codex, but it is dissapointing so maybe it is better with kimi-cli, but I need benchmarks

[xxchan]

codex is optimized for OpenAI responses API, and may not support interleved thinking for kimi-k2-thinking, so it may behave poorly. Kimi CLI should work better.

[lychees]

For tbench, you can found the benchmark result in our tech report.