AI Evolution Tree · A living tree of ML/LLM/AI advancement

Llama nemotron.super.49b

Lab Nvidia

Released Apr 2026

Ling-2.5-1T / Ring-2.5-1T

Mistral Small 4 (v26.03)

NVIDIA Cosmos-Predict 2.5

Cost Moderate

Context 200K

Claude 3.7 Sonnet (Reasoning)

Cost Moderate

Context 128K

DeepSeek Coder V2 Lite Instruct

Lab DeepSeek

Released Jan 2026

Context 128K

DeepSeek LLM 67B Chat (V1)

DeepSeek r1 distill qwen

Lab DeepSeek

Released Jan 2026

ERNIE 5.0 Thinking Preview

Grok 4.20 0309 v2 (Reasoning)

Nemotron Cascade 2 30B A3B

Seed-OSS-36B-Instruct

Nano Banana Pro / Gemini 3 Pro Image

Mistral Large 3 (v25.12)

Lab Black Forest Labs

NVIDIA Isaac GR00T N1

Model Context Protocol

Lab Physical Intelligence

Lab Academic / Independent

Chain-of-Thought Prompting (CoT)

Scaling Laws for Neural Language Models (Kaplan et al.)

Deep RL from Human Preferences (Foundational RLHF)

Attention Is All You Need (Transformer)

AlphaGo Defeats Lee Sedol

Generative Adversarial Networks

Lab Université de Montréal

Lab University of Toronto

Released Sep 2012

Backpropagation for Multi-Layer Networks

Lab Cornell Aeronautical Laboratory

Lab UC San Diego / CMU

Released Nov 1958

1958

1986

2013

2014

2015

2018

2019

2020

2021

2022

2023

2024

2025

2026

nvidia-nemotron-nano

Nvidia

Nvidia Nemotron NANO — Series

13.2B · 128k ctx

jamba

AI21 Labs

Jamba 1 — Series

AI21 Labs

94B active · 256k ctx

step

Stepfun

Step-3 — StepFun's Open Vision Frontier

Stepfun

321B total / 38B active MoE · 66k ctx

llama-nemotron-super

Nvidia

Llama nemotron.super.49b — Series

49B · 128k ctx

Llama

Nvidia

lfm

Liquid AI

Lfm2 — Series

Liquid AI

2.3B active · 33k ctx

Google / DeepMind gemma

Google/DeepMind

Gemma 3 — Google's Open-Weight Family

1B / 4B / 12B / 27B dense · 128k ctx

4B active · 8.4B total · 32k ctx

nvidia-nemotron

Nvidia

Nvidia nemotron.3 — Series

12.7B active · 1M ctx

Claude

Anthropic

Claude 1 — Anthropic's First Public Model

undisclosed · 9k ctx

Claude 3 — Opus, Sonnet, Haiku (Tiered Frontier Family)

Claude 3.5 Sonnet — Mid-Tier Surpasses Prior Flagships

Jun API

Claude 3.7 Sonnet — Hybrid Reasoning Model

Feb API

Claude 4 — Opus & Sonnet (Agentic Frontier Family)

Claude Haiku 4.5 — Fast Tier Reasoning Capable

Claude 3.7 Sonnet (Reasoning)

qwen

Alibaba

Qwen 2.5 — Open Frontier-Comparable from Alibaba

0.5B / 1.5B / 3B / 7B / 14B / 32B / 72B; also Qwen2.5-Coder, Qwen2.5-Math variants · 131k ctx

Qwen 3 — Toggleable Thinking, 119 Languages

0.6B / 1.7B / 4B / 8B / 14B / 32B dense + 30B-A3B and 235B-A22B MoE · 1M ctx

granite

IBM

/ DeepMind Gemini

Google/DeepMind

nova

Amazon

GPT

OpenAI

GPT-2 — Language Models are Unsupervised Multitask Learners

1.5B · 1k ctx

GPT-3 — Language Models are Few-Shot Learners

175B · 2k ctx

GPT-4 — OpenAI's First Multimodal Frontier Model

undisclosed · 8k ctx

GPT-4o — Natively Multimodal OpenAI Frontier

GPT Image 1 / 1.5 — OpenAI's Image Frontier

GPT-oss — OpenAI's First Open-Weight Models in 6 Years

GPT-5 — OpenAI's Unified Successor

20B and 120B · 131k ctx

Aug API

undisclosed · 400k ctx

GPT-5.1 — Faster, Warmer GPT-5 Iteration

GPT-5.2 — Long-Horizon Agentic Work (Instant / Thinking / Pro)

undisclosed · 272k ctx

Dec API

GPT-5.3 Instant — Better Everyday Chat (Less “Caveats”)

undisclosed · 400k ctx

undisclosed · 922k ctx

GPT-5.4 — Professional Work Reasoning (Thinking / Pro)

undisclosed · 922k ctx

GPT-5.5 — Agentic “Do the Work” Model (Thinking / Pro)

undisclosed · 922k ctx

voxtral-tts

Mistral AI

Voxtral TTS (v26.03)

leanstral

Mistral AI

Leanstral (v26.03)

mistral-small

Mistral AI

Mistral Small 4 (v26.03)

Suno v3 — Text-to-Music at Consumer Quality

6.5B active · 256k ctx

Suno

Mar Product

Suno

Suno v5 / v5.5 — Studio-Quality AI Music

Mar Product

Suno

Google/DeepMind

word2vec — Distributed Word Representations

~100-300 dims per word × ~3M vocab ≈ 300M-900M

Oct Open

BERT — Bidirectional Encoder Representations from Transformers

110M / 340M · 512 ctx

AlphaFold 2 — Protein Structure Prediction Solved

~93M

GraphCast — AI Weather Forecasting at ECMWF Quality

37M

AlphaFold 3 — All-Atom Biomolecular Prediction

Veo 3 — Google's Native-Audio Video Model

Genie 3 — Real-Time Interactive World Model

Aug Demo

Nano Banana Pro / Gemini 3 Pro Image — 4K Text-Native Image

Lyria 3 — Google's Generative Music + Live Mode

Feb API

Ant Group

Ling-2.5-1T / Ring-2.5-1T — Ant Group's Trillion-Parameter Twins

Ant Group

~1T total MoE · 128k ctx

NVIDIA Cosmos

Nvidia

NVIDIA Cosmos-Predict 2.5 — World Foundation Model for Physical AI

Claude Opus 4.6 — 1M Context, Sustained Multi-Hour Agency

2B and 14B variants

claude-opus

Anthropic

Lightricks

LTX-2 — Lightricks' Open Video + Audio Model

Lightricks

19B

claude-instant

Anthropic

ernie

Baidu

Jun Open

ERNIE 4.5 — Baidu Goes Open

Baidu

Up to 424B total / 47B active MoE; smaller dense variants from 0.3B · 128k ctx

ERNIE 5.0 Thinking Preview

Baidu

128k ctx

doubao-seed-code

ByteDance

seed-oss

ByteDance

Seed-OSS-36B-Instruct

ByteDance

36.2B · 512k ctx

command-a

Cohere

command-r-plus

Cohere

tiny-aya-global

Cohere

deepseek-coder-v

DeepSeek

DeepSeek Coder V2 Lite Instruct

2.4B active · 16B total · 128k ctx

deepseek-llm

DeepSeek

DeepSeek LLM 67B Chat (V1)

7B · 4k ctx

deepseek-r

DeepSeek

DeepSeek R1 — Open-Weight Reasoning Model

671B total / 37B active · 128k ctx

DeepSeek r1 distill qwen — Series

37B active · 128k ctx

deepseek-v

DeepSeek

DeepSeek V3 — 671B Open MoE at $5.5M Training Cost

671B total / 37B active · 131k ctx

DeepSeek V3.2 — Sparse Attention at Scale

671B total / 37B active MoE · 164k ctx

/ DeepMind palm

Google/DeepMind

minimax-m

MiniMax

Oct Open

MiniMax M2 — Lightning-Fast Open Reasoning

230B total / 10B active MoE · 200k ctx

Minimax m1 — Series

45.9B active · 1M ctx

nemotron-cascade

Nvidia

Nemotron Cascade 2 30B A3B

3B active · 262k ctx

gpt-oss

OpenAI

GPT-oss-120B (high)

OpenAI o1 — Inference-Time Reasoning

5.1B active · 131k ctx

OpenAI

Sep API

OpenAI o3 — Frontier Reasoning at Scale

Grok 3 — xAI's Frontier Bet

Perplexity

37B active · 128k ctx

sonar-reasoning-pro

Perplexity

reka-flash

Reka AI

arctic-instruct

Snowflake

grok

xAI

Feb API

Grok 4 — xAI's Reasoning Frontier

undisclosed · 1M ctx

Aug API

undisclosed · 256k ctx

Grok 4.1 — Hallucination Halved, 2M Context

undisclosed · 2M ctx

Grok 4.20 0309 v2 (Reasoning)

Runway Gen-4.5 — Video Arena Leader

2M ctx

grok-beta

xAI

grok-code-fast

xAI

mimo-v

Xiaomi

ocr

Mistral AI

Runway

Runway

mistral-large

Mistral AI

Mistral Large 3 (v25.12)

41B active · 256k ctx

FLUX

Black Forest Labs

FLUX.2 — Black Forest Labs' Open-Weight Image Frontier

Black Forest Labs

5 variants — klein 4B/9B, schnell, dev 32B, pro

Meta AI

LLaMA — Open Foundation Language Models from Meta

7B / 13B / 33B / 65B · 2k ctx

Llama 2 — Open-Weight LLM with Commercial License

7B / 13B / 70B · 4k ctx

Llama 3 — Meta's Frontier-Comparable Open Model

ESM-3 — Generative Protein Language Model

8B / 70B · 8k ctx

Jun Open

1.4B / 7B / 98B

DINOv3 — Meta's Self-Supervised Vision Foundation

1.7B

SAM 3 — Promptable Concept Segmentation

Gemini 1 — Google's Native Multimodal Frontier

Google Gemini

Google/DeepMind

Dec API

undisclosed · 33k ctx

Gemini Robotics — VLA from DeepMind

Gemini 2.5 Pro — Google's Native Reasoning Frontier

undisclosed · 1M ctx

Gemini 3 Pro — Native Omni-Modal Frontier

undisclosed; MoE · 1M ctx

claude-sonnet

Anthropic

Claude Sonnet 4.6 — The Production Workhorse Update

Marble

World Labs

World Labs Marble — Spatially-Intelligent 3D World Generator

World Labs

Tencent

Hunyuan-Large — Tencent's First Open Frontier MoE

Tencent

389B total / 52B active MoE · 256k ctx

HunyuanVideo 1.5 — Open-Source Video on a 4090

Tencent

8.3B

Moonshot Kimi

Moonshot AI

Kimi Chat — Moonshot AI's Long-Context Bet

Moonshot AI

Kimi K2 — Moonshot's Open MoE Reasoning

Moonshot AI

~1T total / ~32B active · 200k ctx

Kimi K2 Thinking — 200-Tool-Call Agent

Moonshot AI

1T total / 32B active MoE · 256k ctx

MiniMax

MiniMax-Text-01 — Lightning Attention at 4M Context

456B total / 45.9B active · 4M ctx

MiniMax Hailuo 02 — Realtime 1080p Video

Sora — OpenAI's Text-to-Video

Sora

OpenAI

Feb Demo

Sora 2 — Native Audio + Improved Physics

Sep Product

Qwen

Alibaba

Qwen-7B / Qwen-14B — Alibaba's First Major Open Model

7B / 14B · 8k ctx

Qwen3-Coder — Open-Source Coding Agent Frontier

480B-A35B MoE flagship + 30B-A3B Flash · 1M ctx

Qwen3-VL — Open Vision-Language Frontier

2B / 4B / 8B / 32B dense + 30B-A3B and 235B-A22B MoE · 256k ctx

Alibaba

Wan 2.x — Alibaba's Open-Source Video Family

14B

ByteDance

Doubao 1.5 Pro — ByteDance's Sparse MoE Flagship

ByteDance

Seedream 4.0 — ByteDance's MoE Image Speed King

Aug API

ByteDance

MoE

Zhipu

ChatGLM-6B — Tsinghua's Bilingual Open Chat Model

Zhipu

6B · 2k ctx

GLM-4.5 — Zhipu's Open Frontier Family

Zhipu

355B total / 32B active MoE; 106B vision · 128k ctx

ElevenLabs

Jun API

ElevenLabs Eleven v3 — Emotive TTS at Scale

ElevenLabs

Cartesia

Cartesia Sonic — Sub-100ms TTS

Cartesia

Nvidia

NVIDIA Isaac GR00T N1 — Open Humanoid Foundation

2.2B

NVIDIA Parakeet-TDT — Open ASR Leader

0.6B; 1.1B RNNT/CTC; 2.5B Canary-Qwen

Cohere

Cohere Command R+ — RAG-Optimized Open Frontier

Cohere

104B · 128k ctx

Cohere Embed v4 — 128K Context Multimodal Embeddings

Cohere

Kling 2.0 — Kuaishou's Text-to-Video Flagship

Kuaishou

Apr Product

Kuaishou

Meta Llama

Meta AI

Llama 4 — Meta's MoE Open Frontier

Scout 109B/17B active · Maverick 400B/17B active · Behemoth ~2T · 10M ctx

Midjourney

Apr Product

Midjourney V7 — Aesthetic Frontier

Midjourney

Figure Helix — Humanoid VLA

Figure AI

Feb Product

Figure AI

Arc Institute

Evo 2 — DNA Foundation Model at 1Mb Context

Arc Institute

40B · 1M ctx

Microsoft

ResNet — Deep Residual Learning

ResNet-50: 25M · ResNet-152: 60M · trained variants up to ResNet-1001

MatterGen — Generative Materials Discovery

Mistral AI

Mixtral 8x7B — Open Sparse Mixture-of-Experts

46.7B total / 12.9B active per forward pass · 33k ctx

Mistral Large 2 — France's Closed-Frontier Flagship

123B dense · 128k ctx

Pixtral Large — Mistral's 124B Multimodal Frontier

124B · 128k ctx

Codestral 25.01 — Mistral's Coding Model

22B · 256k ctx

phi

Microsoft

Phi-3 — Microsoft's Small Models, Big Quality

Mini 3.8B / Small 7B / Medium 14B · 128k ctx

Phi-4 — Microsoft's Synthetic-Data Small Model

14B dense · 16k ctx

QwQ

Alibaba

QwQ-32B Preview — Alibaba's Open Reasoning Model

32B · 33k ctx

Allen AI

Molmo — Allen AI's Open Vision-Language Model

Allen AI

1B / 7B / 7B / 72B · 4k ctx

OLMo 2 — Allen AI's Fully Open Foundation

Allen AI

7B / 13B / 32B dense · 4k ctx

Anthropic

Model Context Protocol — Open Tool / Context Standard

π₀ — Physical Intelligence's Open Robot Foundation

Physical Intelligence

Oct Open

Physical Intelligence

Perplexity

Sep API

Perplexity Sonar — Search-Native Generation

Perplexity

Sonar Small 8B / Sonar Large 70B / Sonar Huge undisclosed · 128k ctx

DeepSeek

May Open

DeepSeek-V2 — Open MoE with Multi-Head Latent Attention

236B total / 21B active · 128k ctx

Jun Open

DeepSeek-Coder V2 — Open Coding Frontier

236B total / 21B active MoE; 16B and 6.7B sibling sizes · 128k ctx

Stability AI

Stable Diffusion — Open-Weight Latent Diffusion for Images

Stability AI

~890M + ~123M + ~83M

Stable Audio 2.0 — Open Latent-Diffusion Audio

Stability AI

~1.1B

Databricks

DBRX — Databricks' Open MoE

Databricks

132B total / 36B active · 33k ctx

Cognition

Mar Product

Devin — First Autonomous Software Engineer Agent

Cognition

harness over frontier LLMs

BAAI

BGE-M3 — Open Multilingual Multifunction Embeddings

BAAI

568M · 8k ctx

01.AI

Yi-34B — 01.AI's Frontier Open Bilingual Model

01.AI

6B / 34B · 4k ctx

Baidu

Ernie 4.0 — Baidu's Closed Frontier Bet

Baidu

undisclosed · 8k ctx

Academic / Independent

RFdiffusion — Generative Protein Design

Academic / Independent

OpenAI

CLIP — Contrastive Language-Image Pretraining

~400M

InstructGPT — Training LMs to Follow Instructions with Human Feedback

1.3B / 6B / 175B · 2k ctx

Whisper — Open Multilingual Speech Recognition

tiny 39M / base 74M / small 244M / medium 769M / large 1.55B

Nov Product

ChatGPT — RLHF-Aligned Conversational LLM