Moonshot AI's first update, the new Kimi K3.1 Pro > Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0, 0.7 points behind Claude Opus 5.5 and 7.5 ahead...
Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0, just 0.7 points behind Claude Opus 5.5 and 7.5 points ahead of GPT-6 Astra, with a lower average cost per task than either rival's top run. Moonshot also says it filtered cyber-related data out of mid-training and RL as exploit ability grew with post-training, which cut its internal xdev-bench-hard pass@3 from 90.5% to 81.0% — still well above K3's 33.3%. Why it matters: Moonshot's newest open-weight model looks near the frontier on terminal tasks at lower cost, and the company is publicly disclosing the capability price of its safety filtering.
SOURCES
- Rohan Paul2026-10-23 19:21