Reasoning, code, research, multimodal creation, extraction, or high-volume routing.
16 families · no single leaderboard
The model map.
Model names move fast. The durable question is what a family is built for, how you can access it, and where it fits in a real workflow. Compare the ecosystems—not just the launch benchmark.
Hosted product, developer API, open weights, private cloud, or a local device.
Use your own tasks, latency budget, failure modes, and data boundary—not one public score.
Explore the model layer
Find a family by fit.
Versions, prices, limits, and licenses change quickly. Cards describe the durable family shape and link to the provider for current details.
16 entriesHow this map works ↓
01 · Major families
The primary model map.
Eight ecosystems that define much of the current conversation around general-purpose, reasoning, coding, and multimodal AI.
GPTOpenAI
GPT
General-purpose model platform
A broad hosted family spanning fast interaction, deeper reasoning, coding agents, multimodal input, voice, and tool-using workflows.
Strong fitGeneral + agentic work
AccessApps + hosted API
CLDAnthropic
Claude
Long-form reasoning and agents
A hosted model family known for document-heavy work, software engineering, careful writing, and long-running tool-using tasks.
Strong fitCode + knowledge work
AccessApps + hosted API
GEMGoogle
Gemini
Native multimodal model family
Google’s hosted family covers reasoning, fast high-volume work, live interaction, and models that operate across text, images, audio, and video.
Strong fitMultimodal workflows
AccessApps, API + cloud
GRKxAI
Grok
Assistant and real-time tool stack
xAI’s hosted family combines general chat and reasoning with coding, search-connected tools, voice, and generative media surfaces.
Strong fitSearch + interaction
AccessApps + hosted API
DSDeepSeek
DeepSeek
Open reasoning and coding family
A model ecosystem centered on reasoning and software work, distributed through a hosted API and openly available model releases.
Strong fitReasoning + code
AccessAPI + open weights
QWNAlibaba
Qwen
Broad open model ecosystem
A large multilingual family with open releases across general language, code, vision, audio, and specialized deployment sizes.
Strong fitBroad deployment range
AccessAPI + open weights
K3Moonshot AI
Kimi K3
Agentic multimodal frontier model
Moonshot AI’s newly announced Kimi generation targets coding, tool use, multimodal work, and very long inputs. Treat early claims as provisional.
Strong fitCode + long-context agents
Access · 17 JulAPI; weights announced
LMAMeta
Llama
Open-weight ecosystem anchor
Meta’s downloadable model family supports local use, adaptation, research, and a large ecosystem of community tools and derivatives.
Strong fitCustomization + local use
AccessOpen weights
02 · Open, regional & specialist families
The wider model stack.
Important families for private deployment, efficient inference, enterprise retrieval, regional ecosystems, and specialized model building.
MSTMistral AI
Mistral
Efficient open and hosted models
A European model family spanning downloadable releases, hosted frontier models, coding systems, and enterprise deployment options.
Strong fitFlexible deployment
AccessAPI + open weights
GLMZ.ai
GLM
Open and hosted agent models
Z.ai’s family spans general reasoning, software work, multimodal tasks, and open releases with a strong Chinese and global developer footprint.
Strong fitAgents + software work
AccessAPI + open releases
MMXMiniMax
MiniMax
Multimodal and agentic family
A broad model ecosystem covering language, coding, speech, music, image, and video, with both hosted products and open releases.
Strong fitMedia + agent workflows
AccessAPI + open releases
CMDCohere
Command
Enterprise retrieval models
Cohere’s generation family is designed around business applications, retrieval-augmented generation, multilingual work, and controlled deployment.
Strong fitRAG + enterprise data
AccessAPI + private options
NVAAmazon
Nova
Cloud-native model family
Amazon’s Bedrock-integrated family covers text, multimodal understanding, generation, and enterprise workloads inside the AWS stack.
Strong fitAWS-native production
AccessHosted via Bedrock
PHIMicrosoft
Phi
Small language model family
Compact open models designed for efficient reasoning and language workloads where device footprint and deployment control matter.
Strong fitLocal + edge inference
AccessOpen weights + cloud
NEMNVIDIA
Nemotron
Open enterprise model stack
NVIDIA’s model family and data recipes focus on reasoning, agentic work, synthetic-data pipelines, and deployment on accelerated infrastructure.
Strong fitModel building + agents
AccessOpen models + services
ERNBaidu
ERNIE
Knowledge-enhanced model family
Baidu’s model ecosystem combines language and multimodal systems with deep integration across search, cloud, and Chinese-language products.
Strong fitChinese-language stack
AccessHosted + open releases
No model family matches that view.Try another workload or clear the search.
Fit beats hype
How this map works.
Family before version
Cards describe durable provider ecosystems. Use the official catalog link to confirm current model IDs, limits, modalities, and retirement dates.
Your eval before a leaderboard
Public benchmarks are useful clues, not production proof. Test representative tasks, tool failures, latency, cost, and review burden.
Access is part of capability
Hosted models, open weights, and private deployments create different data, licensing, operational, and hardware boundaries.