Atlas

Models

← All models

Kimi K2 Thinking

Moonshot AI · 2025-11-06 · 1T parameters

Kimi K2 Thinking is Moonshot AI's open-weight thinking-agent checkpoint released in November 2025. It is a one-trillion-parameter MoE with 32B active parameters, a 256K-token context window, native step-by-step reasoning interleaved with tool calls, and an INT4 checkpoint produced with quantization-aware post-training.

Benchmark scores

BenchmarkScore
AA-LCR66.3
AA-LCR v1.173.3
AA-Omniscience Index-21.4
AdvancedIF68.5
AdvancedIF (rubric mean score)92.4
Agent Red Teaming k=100 (CAIS)92.8
AIME 2024 (avg@8)85.8
AIME 2024-202585.4
AIME 2025 (Artificial Analysis avg@10)94.7
AIME 2025 (avg@8)85.0
AIME 2025 (MathArena repeated-run mean)92.5
ALE-Bench1103
Apex (MathArena)0.0
APEX (Mercor)6.0
Apex Shortlist47.3
APEX-SWE17.4
APEX-v1-extended52.4
Arena Code WebDev — Brand & Marketing1335
Arena Code WebDev — Consumer Product1320
Arena Code WebDev — Content Creation Tools1324
Arena Code WebDev — Data & Analytics1322
Arena Code WebDev — Frontend1321
Arena Code WebDev — Gaming1312
Arena Code WebDev — HTML1325
Arena Code WebDev — React1315
Arena Code WebDev — Reference-Based Design1314
Arena Code WebDev — Simulations1301
Arena Text — Business, Management & Financial Ops (No Style Control)1397
Arena Text — Chinese (No Style Control)1456
Arena Text — Creative Writing (No Style Control)1373
Loading Atlas data…