PGP

Moonshot AI
2026-07-17 17:36:42

Moonshot’s Kimi K3 tops several AI benchmarks, with Claude Fable 5 still ahead on composite score

Moonshot AI has released Kimi K3, which Decrypt described as the largest Chinese open-source model yet made public. The model carries 2.8 trillion parameters in a mixture-of-experts design and was reported to have moved to the top of several benchmark rankings, including Towards AI’s Writing Elo and Arena AI’s Frontend Code Leaderboard. In those tests, K3 posted a 2,840 Writing Elo score versus Claude Fable 5’s 2,760, and 1,679 on the frontend coding board versus Fable 5’s 1,631. The broader picture is more mixed. On the Artificial Analysis Intelligence Index, which aggregates nine independent evaluations, K3 scored 57. Claude Fable 5 scored 60, GPT-5.6 Sol 59, and Claude Opus 4.8 56, leaving K3 in third place on that composite. Decrypt also cited BridgeBench and a web engineering benchmark where K3 was reported ahead of Fable in several head-to-head measures. Moonshot said K3 supports a 1 million-token context window, native image and video understanding, and “always-on reasoning.” The company plans to release weights on July 27 for large enterprises and businesses. The report also noted a rise in hallucination rate versus K2.6, from 39% to 51%, and said heavy traffic has made the free web version difficult to use consistently.

1600
Moonshot’s Kimi K3 tops several AI benchmarks, with Claude Fable 5 still ahead on composite score