CLAIM · ASSESSED ~ · CONFIDENCE 0.55
Kimi K3 scored 88.3 on Terminal Bench 2.1, second to GPT 5.6 Sol's 88.8 and above Claude Fable 5's 84.6.
Part of the monitored dynamic Global AI Governance Race · VUCA INDEX 69/100
EVIDENCE CHAIN · 1
INGESTED ARTICLETIER 3JUL 17
ithome.com.tw — Moonshot AI發表Kimi K3，效能逼近Claude與GPT最新模型
"Kimi K3於評估終端工程任務能力的Terminal Bench 2.1取得88.3分，僅次於GPT 5.6 Sol的88.8分，並高於Claude Fable 5與Claude Opus 4.8的84.6分"
ENTITIES
PROVENANCE
Extracted by pipeline v0.5 (claude-opus-4-8) from ithome.com.tw · approved by christopher@vucanews.com JUL 18.
Extracted JUL 17; approved JUL 18 at 0.55.
MORE FROM THIS DYNAMIC
- Anthropic national security head Tarun Chhabra said the US holds a six-to-nine-month lead in frontier AI models over Chinese competitors.
- Chhabra accused Chinese labs of using unauthorised model 'distillation' to extract IP from US models, calling the practice adversarial.
- Anthropic is shutting down roughly millions of accounts per week that facilitate model distillation.
- Moonshot AI released Kimi K3 on 2026-07-16, a mixture-of-experts model with 2.8 trillion total parameters, claimed as the world's largest open model.
- Moonshot AI plans to release full Kimi K3 model weights on 2026-07-27.
STRUCTURED DISSENT