First Chinese, open model that doesn’t feel like benchmaxing a distillation attack.
On Basecamp Bench BE, Kimi K3 scored second highest and is better than Grok 4.5 and Sol 5.6 on several dimensions.
K3 was 50% cheaper than Sol to run and 2.25x slower than Fable (the slowest), however API outages prevented completion of the FE test.
Kimi K3 Basecamp Bench BE scores across nine dimensions