Tip

Moonshot AI Kimi K3 Model AI Agent Knowledge Work Score Surpasses GPT-5.6 Sol, Second Only to Claude Fable 5

Published: 2026-07-24 14:33 Source: IT Home (RSS)

Moonshot AI's Kimi K3 model scored 1543 points on the AA-Briefcase benchmark, surpassing GPT-5.6 Sol (1501 points) and second only to Claude Fable 5 (1574 points). The model has 2.8 trillion parameters and a 1 million token context window. Input costs are 2 yuan per million tokens with cache hit and 20 yuan without, while output costs 100 yuan.

Read the original (opens in a new tab)

News stream data aggregated by AI HOT

Related newsLatest in this category

Tip Tibo pauses X updates, returning tomorrow

07-28 13:44 · X: Tibo (@thsottiaux)

Tip Kimi K3 Technical Report: 2.8 Trillion Parameters, 2.5x Compute Efficiency Improvement

07-28 13:29 · X: Xiaohu (@xiaohu)

Tip GLM 700B Parameter Model Performance Approaches Opus Level

07-28 13:26 · X: Aravind Srinivas (Perplexity CEO) (@AravSrinivas)

Tip Kimi K3 open source ignites AI open source war, Huang Renxun joins 77 companies in petition against sanctions

07-28 12:59 · X: Kazik (@Khazix0918)

Tip From Self-Developed JIRA to Returning to Linear: The Maintenance Dilemma of Vibe Coding

07-28 12:56 · X: Oran Ge (@oran_ge)