Google DeepMindLab Rank #01 · 82.2 / 100 ·
Gemini
Flagship models; read text, images, audio and video
3.8
3.7
3.6
Model Atlas
Google DeepMind
OpenAIOpenAI
Lab Rank #02 · 82.1 / 100 ·
GPT
Flagship chat models, from cheap Nano to costly Pro
6.1
6
5.6
QwenQwen
Lab Rank #03 · 77.6 / 100 ·
Qwen
Flagship chat models, from small to very large
3.8
3.7
3.5
AnthropicAnthropic
Lab Rank #04 · 72.6 / 100 ·
Opus
Flagship model; the lab's top Intelligence Index score
5.5
5
4.8
MetaMeta
Lab Rank #05 · 68.7 / 100 ·
Muse
Flagship multimodal reasoning for agents and coding
1.3
1.2
30B
xAIxAI
Lab Rank #06 · 68.0 / 100 ·
Grok
Flagship reasoning models for coding and agentic work
4.7
4.6
4.5
XiaomiXiaomi
Lab Rank #07 · 67.4 / 100 ·
Mimo
V2.6
V2.5
V2
Moonshot AIMoonshot AI
Lab Rank #08 · 63.3 / 100 ·
Kimi
0905
DeepSeekDeepSeek
Lab Rank #09 · 63.0 / 100 ·
DeepSeek
Flagship chat models, with cheap Flash and larger Pro
V4.1
V4
V3.2
Z.aiZ.ai
Lab Rank #10 · 62.2 / 100 ·
GLM
5.3
5.2
5.1
MiniMaxMiniMax
Lab Rank #11 · 57.9 / 100 ·
Minimax
Language models for coding and agentic work
01
Mistral AIMistral AI
Lab Rank #12 · 52.7 / 100 ·
Mistral
Flagship chat models in Small, Medium and Large
4
3
2603
MicrosoftMicrosoft
Lab Rank #13 · 52.1 / 100 ·
PHI
Small model for reasoning with limited memory
4
NVIDIANVIDIA
Lab Rank #14 · 51.2 / 100 ·
Nemotron
Open reasoning models, from Nano to Ultra
3.5
3
TencentTencent
Lab Rank #15 · 50.5 / 100 ·
HY3
Agentic model with configurable reasoning levels
CohereCohere
Lab Rank #16 · 45.6 / 100 ·
Command
Flagship chat models for RAG, tools and enterprise agents
A
R7b
R
Lab Rank methodology+−
Benchmark standing uses independently published results that share the same protocol, scope, unit, and evidence date. The best comparable release represents each lab in a benchmark run.
A lab needs at least five comparable runs and three canonical releases to rank. Missing evidence stays missing; scores are not inherited across a model family. This measures represented Atlas evidence, not company quality or importance.