Moonshot AI Kimi K3 Ranks Third in Intelligence Index With Performance Rivaling Top Models
Moonshot AI's new Kimi K3 model secures the third position on the Artificial Analysis Intelligence Index with performance competitive with top-tier industry models.
1. Performance and Intelligence Rankings
Moonshot AI’s new Kimi K3 model has achieved a score of 57 on the Artificial Analysis Intelligence Index, securing the number three position. This performance places it in a tier comparable to Opus 4.8 and GPT-5.5, though it remains behind Claude Fable 5 and GPT-5.6 Sol. The model demonstrates significant improvements in agentic task performance, reaching an Elo rating of 1668 on GDPval-AA v2 and securing the top spot on AutomationBench-AA. Additionally, Kimi K3 ranks second in agentic knowledge work, trailing only Claude Fable 5 in the AA-Briefcase evaluation.
2. Technical Specifications and Efficiency
Kimi K3 is a 2.8T parameter model featuring a 1M token context window and native support for multimodal text and image inputs. Despite its increased size compared to the 1T parameter Kimi K2.6, the new model exhibits higher token efficiency, utilizing 21% fewer output tokens to complete evaluations while achieving higher scores. While the model currently shows improved accuracy rates, data indicates a regression in hallucination rates, which rose from 39% in the previous version to 51%.
3. Pricing and Future Availability
The Kimi K3 API is priced at $3.00 per 1M input tokens and $15.00 per 1M output tokens, with a 90% discount available for cached inputs. On a cost-per-task basis, Kimi K3 averages $0.94, making it more cost-effective than Opus 4.8 ($1.80) and comparable to GPT-5.6 Sol ($1.04). Although the model is currently accessible only through Moonshot AI’s first-party API, the company has announced plans to release the model weights. Once released, Kimi K3 is expected to become the leading open weights model, significantly outpacing current alternatives like GLM-5.2 and DeepSeek v4 Pro.