Atlas

Models

← All models

Kimi K2.6

Moonshot AI · 2026-04-20 · 1T parameters

Kimi K2.6 is Moonshot AI's open-weight, native-multimodal Mixture-of-Experts checkpoint released on April 20, 2026, with a 262,144-token context window and optional Thinking and Instant modes. It retains Kimi K2.5's architecture, pairing a 61-layer MLA language model with MoonViT, and ships with native INT4 quantization for text, image, and video input.

Benchmark scores

BenchmarkScore
A-Fantasia60.0
A-Fantasia (Valid Attempts with Retries)56.0
A-Fantasia Backwards Spelling70.0
A-Fantasia Chess29.0
A-Fantasia Chess (Valid Attempts with Retries)32.0
A-Fantasia Cube (Valid Attempts with Retries)64.0
A-Fantasia Cube Rotation81.0
A-Fantasia Spelling (Valid Attempts with Retries)72.0
AA-Briefcase Elo824
AA-Briefcase v1.1 Elo820
AA-LCR69.7
AA-LCR v1.181.0
AA-Omniscience Index5.3
Agent Red Teaming k=100 (CAIS)90.7
Agents' Last Exam (ALE)9.2
Agents' Last Exam ALE-CLI Pass Rate8.1
Agents' Last Exam ALE-CLI Score21.2
Agents' Last Exam Full-Spectrum Pass Rate6.4
Agents' Last Exam Full-Spectrum Score18.2
Agents' Last Exam Last-Exam Pass Rate0.0
Agents' Last Exam Last-Exam Score1.6
Agents' Last Exam Near-Term Pass Rate15.7
Agents' Last Exam Near-Term Score35.1
Agents' Last Exam Overall Score21.7
Agents' Last Exam Unlicensed Full-Spectrum Pass Rate7.0
Agents' Last Exam Unlicensed Full-Spectrum Score18.8
Agents' Last Exam Unlicensed Last-Exam Pass Rate0.0
Agents' Last Exam Unlicensed Last-Exam Score2.1
Agents' Last Exam Unlicensed Near-Term Pass Rate15.7
Agents' Last Exam Unlicensed Near-Term Score35.1
Loading Atlas data…