18+Agents Tracked
74.8%Top SWE-bench
~$1.15Avg Task Cost
62%OSS & BYOK Ratio
Methodology

βš–οΈ Agent Leaderboard Methodology & Scoring Fairness

Coding agents (Agent Harnesses) are evaluated on end-to-end engineering autonomy rather than raw next-token prediction:

40% weight

SWE-bench Verified (40%)

End-to-end resolution of real GitHub issues passing test assertions

25% weight

Harness & Sandboxing (25%)

Terminal Bash execution, Git Worktree isolation & reactive subagents

20% weight

Cost & Speed Efficiency (20%)

Token dollar expense and total wall-clock time per completed task

15% weight

Open Ecosystem & MCP (15%)

Custom MCP tool support, BYOK (Bring Your Own Key) & model routing

Note: Scores reflect holistic engineering autonomy, stability, and ROI, rather than single-turn model intelligence alone.

Pricing Model:
Supported LLM:
#1
Championcli

Google Antigravity (AGY)

Google DeepMindβ˜… 28.4K stars

Google DeepMind frontier agentic coding system featuring reactive wakeup subagents and Git worktree isolation for end-to-end multi-file refactoring and MCP extension.

SWE-bench Verified74.8%
Cost / Task$1.15
Avg Time3.8m
Pricing ModelOpen Ecosystem (BYOK)
Gemini 3.7 ProGemini 3.7 FlashDeepSeek-V4Claude 3.7 Sonnet
βœ“ Reactive subagents βœ“ Git worktree isolation βœ“ Automated unit-test fixing βœ“ MCP ecosystem βœ“ Multi-file coordinated edits
#2
Cloud Autonomouscloud

Devin 2.0 (Cognition)

Cognition AI

Cognition AI first autonomous cloud AI software engineer equipped with isolated multi-VM Linux sandboxes and real-time browser debugging.

SWE-bench Verified71.5%
Cost / Task$4.20
Avg Time8.5m
Pricing ModelSubscription ($500/mo) / ACU
GPT-5.6 SolClaude 3.7 Sonneto3-mini
βœ“ Isolated multi-VM cloud sandboxes βœ“ Live browser hot-reload debugging βœ“ Deep Slack / Jira / Linear integration βœ“ End-to-end PR delivery
#3
Official CLIcli

Claude Code (Anthropic)

Anthropicβ˜… 18.2K stars

Anthropic official agentic coding CLI tool running natively in your terminal with adaptive reasoning on Claude Sonnet 5 and Opus 4.8.

SWE-bench Verified70.3%
Cost / Task$1.85
Avg Time4.2m
Pricing ModelBYOK (Anthropic API)
Claude Sonnet 5Claude Opus 4.8Claude 3.7 SonnetClaude 3.5 Sonnet
βœ“ Fast local terminal runs βœ“ Adaptive thinking token budget βœ“ Semantic codebase indexing βœ“ One-step structured commits
#4
Top IDEide

Cursor Composer (Agent Mode)

Anysphere

AI-first IDE Cursor native Composer agent with background shadow workspace linting and multi-model routing.

SWE-bench Verified68.2%
Cost / Task$0.65
Avg Time2.1m
Pricing ModelFreemium / Pro ($20/mo)
GPT-5.6 SolDeepSeek-V4Claude 3.7 Sonneto3-mini
βœ“ Background shadow-workspace builds βœ“ Fast Tab completion βœ“ Multi-model routing βœ“ Deep Rules alignment
#5
Flow IDEide

Windsurf Cascade (Flows)

Codeium

Codeium next-gen agentic IDE with Cascade Flows paradigm for human-in-the-loop intent sync and deep codebase indexing.

SWE-bench Verified66.4%
Cost / Task$0.55
Avg Time2.4m
Pricing ModelFreemium / Pro ($15/mo)
GPT-5.6 SolDeepSeek-V4Claude 3.7 Sonnet
βœ“ Cascade Flows intent orchestration βœ“ Fast global codebase indexing βœ“ Native terminal shell interaction βœ“ Human-in-the-loop interrupt and steer
#6
Enterprise Leaderide

GitHub Copilot (Agent Mode / Codex)

GitHub / OpenAI

GitHub and OpenAI official coding agent supporting VS Code Agent mode, Copilot Workspace, autonomous issue remediation, and enterprise compliance.

SWE-bench Verified66.8%
Cost / Task$0.45
Avg Time2.2m
Pricing ModelSubscription ($10/mo)
GPT-5.6 SolGemini 3.7 ProClaude 3.7 Sonneto3-mini
βœ“ Native VS Code agent mode βœ“ Copilot Workspace end-to-end delivery βœ“ Automated GitHub issue fixes and PRs βœ“ Cross-repo semantic indexing with enterprise compliance
#7
Top OSScli

OpenHands (All-Hands AI)

All-Hands AIβ˜… 42.5K stars

Fully open-source AI software engineer platform with Docker container isolation and multi-model support.

SWE-bench Verified65.8%
Cost / Task$1.45
Avg Time5.6m
Pricing ModelOpen Source (Apache-2.0)
DeepSeek-V4DeepSeek-R1Qwen2.5-Coder-32BClaude 3.7 Sonnet
βœ“ Fully open source, no lock-in βœ“ Docker container sandboxing βœ“ Self-hosted with local Ollama βœ“ Visual web console
#8
Top CLIcli

Aider

Paul Gauthierβ˜… 31.2K stars

Command line AI pair programmer with Tree-sitter repo map, automatic Git commits, and extreme token efficiency.

SWE-bench Verified61.2%
Cost / Task$0.45
Avg Time1.8m
Pricing ModelOpen Source (Apache-2.0)
DeepSeek-V4GPT-5.6Gemini 3.7Local LLMs
βœ“ Git-aware Commits βœ“ Repo Map Tree-sitter βœ“ Universal CLI βœ“ Architect/Editor Mode βœ“ Voice Mode
#9
Top Extensionextension

Cline (VS Code Agent)

Cline Open Sourceβ˜… 24.8K stars

Top open-source VS Code autonomous agent with native MCP tool server integration and Sequential Thinking.

SWE-bench Verified64.5%
Cost / Task$0.75
Avg Time3.2m
Pricing ModelOpen Source (Apache-2.0)
DeepSeek-V4Gemini 3.7 ProClaude 3.7 Sonneto3-mini
βœ“ Dynamic MCP toolchain mounting βœ“ Multi-step sequential thinking βœ“ Bring your own API key, no markup βœ“ Terminal commands with approval
#10
Fullstack Webcloud

Bolt.new

StackBlitz

In-browser AI web development agent running fullstack Node.js environments via WebContainers with 1-click deployment.

SWE-bench Verified58.5%
Cost / Task$1.10
Avg Time1.5m
Pricing ModelFreemium / Token Subscription
DeepSeek-V4GPT-5.6Claude 3.7
βœ“ WebContainers in-browser Node βœ“ 1-Click Fullstack Deploy βœ“ Instant Hot-reload βœ“ Netlify/Supabase
#11
UI & Fullstackcloud

Lovable

Lovable Dev

GPT-engineer inspired autonomous web app builder creating production React apps with Supabase backends.

SWE-bench Verified57.2%
Cost / Task$1.30
Avg Time1.6m
Pricing ModelSubscription ($20/mo)
GPT-5.6Claude 3.7
βœ“ React + Tailwind Gen βœ“ Supabase Backend Integration βœ“ GitHub Sync βœ“ Visual UI Editor
#12
New IDEide

Trae (ByteDance)

ByteDance

ByteDance native AI IDE with Builder mode, natively offering Claude 3.7 and DeepSeek-V3 for fullstack coding.

SWE-bench Verified64.9%
Cost / Task$0.40
Avg Time2.8m
Pricing ModelFree Tier Available
DeepSeek-V4DeepSeek-R1Doubao-Seed-2.0-CodeClaude 3.7 Sonnet
βœ“ Builder full-stack autonomous builds βœ“ Free built-in access to top models βœ“ Optimized for Chinese-language development βœ“ Smart terminal interaction
#13
Design to Codecloud

v0 by Vercel

Vercel

Generative UI system by Vercel producing accessible React and shadcn/ui components from natural language prompts.

SWE-bench Verified55.8%
Cost / Task$0.80
Avg Time1.2m
Pricing ModelFreemium / Premium ($20/mo)
GPT-5.6Custom Vercel ModelsClaude 3.7
βœ“ shadcn/ui + Tailwind βœ“ React RSC Components βœ“ Next.js Integration βœ“ Figma Import
#14
Personal AIcloud

Pi Agent (Inflection)

Inflection AI

Pioneering personal intelligence agent by Inflection AI designed for empathetic conversation, active listening, and enterprise knowledge synthesis.

SWE-bench Verified55.2%
Cost / Task$1.20
Avg Time2.5m
Pricing ModelInflection Enterprise / Pi Pro
Inflection 3.0Inflection 2.5Pi Intelligence
βœ“ Empathetic Personal Agent βœ“ Real-time Web Search βœ“ Voice & Multimodal βœ“ Enterprise Knowledge Assistant
#15
OSS Extensionextension

Continue

Continue Devβ˜… 22.4K stars

The leading open-source AI code assistant for VS Code and JetBrains, supporting full customization and local LLMs.

SWE-bench Verified54.2%
Cost / Task$0.35
Avg Time2.6m
Pricing ModelOpen Source (Apache-2.0)
OllamaDeepSeek-V4ClaudeOpenAI
βœ“ VS Code & JetBrains βœ“ Custom Slash Commands βœ“ Context Providers βœ“ Enterprise Self-hosted
#16
Academiccli

SWE-agent

Princeton NLPβ˜… 16.5K stars

Princeton benchmark-defining autonomous SWE agent utilizing specialized Agent-Computer Interfaces (ACI).

SWE-bench Verified53.8%
Cost / Task$2.20
Avg Time7.8m
Pricing ModelOpen Source (MIT)
GPT-5.6DeepSeek-V4Claude 3.7
βœ“ Agent-Computer Interface (ACI) βœ“ Docker Isolation βœ“ Benchmark Rigor βœ“ Academic Research
#17
Cloud Platformcloud

Replit Agent

Replit

Autonomous fullstack software developer in Replit cloud environment that scaffolds apps and provisions databases.

SWE-bench Verified52.1%
Cost / Task$1.50
Avg Time3.5m
Pricing ModelReplit Core ($25/mo)
GPT-5.6Claude 3.7
βœ“ Cloud IDE βœ“ PostgreSQL Provisioning βœ“ Deployment Hosting βœ“ Mobile App Coding
#18
Enterprise OSScli

Goose AI

Block / Squareβ˜… 7.8K stars

Open-source on-machine developer agent built by Block to automate engineering tasks from the terminal.

SWE-bench Verified50.8%
Cost / Task$0.85
Avg Time3m
Pricing ModelOpen Source (Apache-2.0)
DeepSeek-V4OpenAIOllamaClaude 3.7
βœ“ Autonomous CLI βœ“ Extensible Extensions βœ“ Desktop App βœ“ MCP Client