.num-cell{font-family:ui-monospace,'SF Mono',monospace;text-align:right}

AI agent frameworks compared — 2026

Data-driven comparison of 11 major AI agent frameworks from the Nerq index of 5M+ assets. Published 2026-08-25.

11
frameworks compared
24,427
total agents using these
Anthropic
most adopted

comparison table

frameworkagentsavg trusttop agentlanguagenew (30d)
Anthropic 7,490 60.7 IBM/mcp-context-forge (86.4) Python 0
OpenAI 6,329 59.9 IBM/mcp-context-forge (86.4) Python 0
LangChain 2,680 59.4 vstorm-co/full-stack-fastapi-n (86.2) Python 0
MCP 2,046 60.6 IBM/mcp-context-forge (86.4) Python 0
Ollama 1,900 60.0 infiniflow/ragflow (82.9) Python 0
HuggingFace 1,238 60.3 Giskard-AI/giskard-oss (85.6) Python 0
AutoGen 1,113 59.4 NVIDIA/NeMo-Agent-Toolkit (81.3) Python 0
CrewAI 809 59.2 vstorm-co/full-stack-fastapi-n (86.2) Python 0
LlamaIndex 465 58.0 topoteretes/cognee (82.5) Python 0
A2A 191 59.7 GoogleCloudPlatform/agent-star (83.5) Python 0
Semantic Kernel 166 57.3 NVIDIA/NeMo-Agent-Toolkit (81.3) Python 0

adoption — agent count

Number of agents in the Nerq index using each framework

Anthropic
7,490
OpenAI
6,329
LangChain
2,680
MCP
2,046
Ollama
1,900
HuggingFace
1,238
AutoGen
1,113
CrewAI
809
LlamaIndex
465
A2A
191
Semantic Kernel
166

average trust score

Mean trust score of agents using each framework (0-100)

Anthropic
61
MCP
61
HuggingFace
60
Ollama
60
OpenAI
60
A2A
60
LangChain
59
AutoGen
59
CrewAI
59
LlamaIndex
58
Semantic Kernel
57

methodology

Framework association is determined from the frameworks array in the Nerq agent database. An agent is counted under a framework if that framework appears in its declared dependencies, metadata, or detected integrations. Trust scores are the Nerq Trust Score v2 — a composite of security practices, compliance, maintenance activity, community trust, and ecosystem compatibility. Data updates hourly from the live index.

related

State of AI Assets Q1 2026 · Best Coding Agents · Benchmark API · Weekly Signal

Data from the Nerq index. JSON: /v1/agent/stats · API docs · all reports

We use cookies for analytics and caching. Privacy Policy