AI Model Timeline

Tracking the accelerating release frequency of frontier AI models.

Kimi K3

Moonshot AI·

LLMsopen-weightFrontier

Overview

2.8T MoE (~50B active, 16 of 896 experts) native multimodal flagship with a 1M-token context window. Open weights released on Hugging Face 2026-07-27 under the bespoke Kimi K3 License — the largest open-weight model at release.

Capabilities and innovations

2.8T total / ~50B active MoE (896 experts, 16 routed)1M token context windowNatively multimodal (text, image, video)Vision-in-the-loop (screenshot inspect + code edit)Kimi K3 License (revenue-tiered)Kimi Delta Attention2.8T Open-Weight MoE (largest open-weight at release)Vision-in-the-Loop Agent Feedback1M-Token Native Context

Benchmarks

BenchmarkScoreSource
GPQA Diamond93.5%vendor
GPQA-Diamond 93.5 (official launch blog, 2026-07-16).
Humanity’s Last Exam (no tools)43.5%vendor
HLE-Full 43.5 no-tools (text-only), recorded for cross-model comparability. Moonshot also reports 56.0 with tools, which is not comparable to the no-tools HLE used here.
Terminal-Bench 2.x88.3%vendor
Terminal-Bench 2.1 88.3 (official launch blog).
MMMU-Pro81.6%vendor
MMMU-Pro 81.6 (official launch blog).
  • SWE-bench Pro: not reported by the vendor — Moonshot reports its own coding suite (Program Bench, SWE Marathon, FrontierSWE, DeepSWE) instead of SWE-bench Pro.
  • MMLU-Pro: not reported by the vendor

Architecture and hardware

Parameters
2800B total, 50B active per token (MoE)
Estimated VRAM at Q4
~1610 GB, Frontier class
Quantization formats
INT4, BF16, GGUF
Recommended runtime
vLLM
License
Kimi K3 License (revenue-tiered)

Links

More from Moonshot AI

Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.