Compare AI Coding Agents

Q: Which AI coding agents support MCP?

60 AI coding agents support MCP (Model Context Protocol): Amp, Ante, Antigravity CLI, Auggie CLI, AutoHand, Blackbox AI CLI, Charm Crush, Claude Code, Cline CLI, Codebuff, Codewhale, Codex CLI, Command Code, Cosine, Cursor CLI, Deep Agents Code, Devin for Terminal, Droid CLI, ForgeCode, fx (by Vercel), Gemini CLI, GitHub Copilot CLI, Goose CLI, Grok Build CLI, Groq Code CLI, IBM Bob, jcode, Jules CLI, Junie CLI, Keen Code, Kilo Code, Kimi Code CLI, Kiro CLI, Letta Code, Mastra Code, MiMo Code, Mistral Vibe, Muse Code, Nanocoder, Neovate Code, omp, OpenBlock, OpenCode, OpenHands CLI, Open Interpreter, Poolside Agent CLI, Prime Agent, Qoder CLI, Qwen Code, Ralph TUI, Rovo Dev CLI, Snowflake Cortex CLI, Standard Code, Toad, Vix, VT Code, Warp Agent CLI, Zero, zerostack, Continue.

Q: Which AI coding agents are open source?

42 AI coding agents are open source: Aider, AutoHand, Blackbox AI CLI, Charm Crush, Cline CLI, Codebuff, Codewhale, Codex CLI, Deep Agents Code, ForgeCode, fx (by Vercel), Gemini CLI, Goose CLI, Grok Build CLI, Groq Code CLI, jcode, Keen Code, Kilo Code, Kimi Code CLI, Letta Code, Mastra Code, MiMo Code, Mistral Vibe, Nanocode, Nanocoder, Neovate Code, omp, OpenCode, OpenHands CLI, Open Interpreter, Pi, Prime Agent, Qwen Code, Ralph TUI, Toad, Vix, VT Code, Zero, zerostack, Zot, Continue, Plandex.

Q: What is the best AI coding assistant?

The best AI coding assistant depends on your workflow and budget. For terminal-based development, agents like Claude Code, Aider and Cursor CLI lead in Terminal Bench accuracy. Use the comparison table to filter by features like MCP support, BYOK, multi-file editing, and pricing to find the best fit.

Compare 65 AI coding agents, also called AI coding assistants, side-by-side. Features, pricing, Terminal Bench 4.0 and Artificial Analysis benchmarks, MCP support, and more.

Looking for an overview of all AI coding agents?

Pick 2 to 4 agents to compare side-by-side or compare from the large table below.

AI Coding Agents Comparison Table - Compare features, pricing, benchmarks, and more across 65 agents.
Agent	Provider	Interface	TBench Rank	TBench Accuracy	AA Rank Coding Agent Index	AA Score Coding Agent Index	Pricing	Free Tier	BYOK	Price Range	MCP	Multi-Agent	Multi-File	Auto Test	Sandbox	Multimodal	Voice	Hooks	Web Search	Local Models	Git Integration	IDE Support	Context Window	License	Language
Aider	Independent	cli					BYOK	✓	✓	Free (BYOK)	✗	✗	✓	✓	✗	✓	✓	✓	✗	✓	deep	VS Code, Neovim, Emacs	Model-dependent	Open Source (Apache-2.0)	python
Amp	AI Lab	cli					Usage Based	✓	✗	Free daily grant, then usage-based	✓	✓	✓	✓	✓	✓	✗	✓	✓	✗	deep	VS Code	Model-dependent	Proprietary Software	typescript
Ante	Independent	both	#23	69.4%			Free, BYOK	✓	✓	Free (alpha preview)	✓	✓	✓	✗	✓	✓	✗	✗	✓	✓	basic	None shipped; `ante serve` exposes a JSONL protocol for editor plugins	Model-dependent	Proprietary Software	rust
Antigravity CLI	AI Lab	both			#8	42%	Freemium, Subscription, Usage Based, BYOK	✓	✓	$0 - $200	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	Antigravity IDE	Model-dependent	Proprietary Software	go
Auggie CLI	Independent	both					Subscription	✓	✗	$100-200/mo	✓	✓	✓	✓	✓	✓	✓	✓	✓	✗	deep	VS Code, JetBrains IDEs (ACP), Zed (ACP)	Not disclosed	Proprietary Software	typescript
AutoHand	Independent	cli					Freemium	✓	✓	Free (BYOK) / Paid (commercial >$5M ARR)	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	VS Code, Zed, JetBrains	Model-dependent; auto-compaction	Open Source (Apache-2.0)	typescript
Blackbox AI CLI	Independent	cli					Freemium	✓	✓	Free (BYOK) / $10-40/mo	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	35+ IDE integrations including VS Code, JetBrains	Model-dependent	Open Source (GPL-3.0)	typescript
Charm Crush	Independent	tui					BYOK	✓	✓	Free (BYOK), $0 - $20 / month	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	None - Terminal only	Model-dependent	Open Source (FSL-1.1-MIT)	go
Claude Code	AI Lab	cli	#47	58.0%	#1	62%	Subscription	✗	✗	$17-200/mo	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	VS Code, JetBrains	1M tokens	Proprietary Software	typescript
Cline CLI	Independent	both					Freemium	✓	✓	Free (BYOK) / $20/mo per user	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	basic	VS Code, JetBrains, Zed, Neovim, Emacs	Model-dependent (up to 1M)	Open Source (Apache-2.0)	typescript
Codebuff	Independent	cli					Freemium	✓	✗	Free 500 credits/mo / $100-500/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	None - Terminal only	Model-dependent	Open Source (Apache-2.0)	typescript
Codewhale	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	Editors via ACP (registry submission pending)	Model-dependent	Open Source (MIT)	rust
Codex CLI	AI Lab	both	#4	82.2%	#3	62%	Subscription	✗	✓	$20-200/mo	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	VS Code, Zed (via ACP)	Model-dependent	Open Source (Apache-2.0)	rust
Command Code	Independent	both					Freemium	✓	✓	$10 free credits / Paid plans	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	Integrates with VSCode based editors	Not disclosed	Proprietary Software	typescript
Cosine	AI Lab	both					Usage Based, Freemium, Subscription	✗	✓	$19-999/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✗	deep	VS Code	Model dependent	Proprietary Software	go
Cursor CLI	Independent	cli					Freemium	✓	✗	Free / $20-200/mo	✓	✓	✓	✓	✓	✓	✓	✓	✓	✗	deep	Cursor IDE, Xcode, Android Studio, JetBrains, VSCode, Neovim	Up to 1M tokens	Proprietary Software	typescript
Deep Agents Code	Independent	both	#26	66.5%			BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✗	✓	✗	✗	✓	✓	✓	basic	None - Terminal only	Model-dependent; compression at threshold	Open Source (MIT)	python
Devin for Terminal	AI Lab	tui			#2	62%	Subscription, Usage Based	✓	✗	$0, $20 - $200 per month	✓	✓	✓	—	✓	✓	—	✓	✓	✗	deep	Windsurf IDE	Model-dependent	Proprietary Software	rust
Droid CLI	Independent	both	#9	77.3%			Freemium	✓	✓	$20-200/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	VS Code, Cursor, Windsurf, JetBrains IDEs, Vim, Zed	Model-dependent	Source Available (Proprietary)	typescript
ForgeCode	Independent	both					BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	VSCode & (terminal-native, ZSH plugin)	Model-dependent	Open Source (Apache-2.0)	rust
fx (by Vercel)	Independent	both					Free, BYOK, Usage Based	✗	✓	0 - $20 / mo / Custom (Vercel AI Gateway)	✓	—	✓	✗	✓	✓	✗	✗	✓	✓	basic	None - Terminal only	Model-dependent	Open Source (Apache-2.0)	zig
Gemini CLI	AI Lab	both	#39	61.4%			Free	✓	✗	Free (1,000 req/day)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	Antigravity, VS Code, JetBrains IDEs (ACP), Zed (ACP)	1M tokens	Open Source (Apache-2.0)	typescript
GitHub Copilot CLI	AI Lab	both					Freemium, Usage Based	✓	✓	Free / $10-39/mo	✓	✓	✓	✓	✗	✓	✓	✓	✓	✓	deep	VS Code, Cursor, Windsurf, JetBrains, Neovim, Visual Studio, Xcode	Auto-managed	Proprietary Software	typescript
Goose CLI	Independent	both	#53	54.3%			BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	basic	Cursor, VSCode, JetBrains IDEs	Model-dependent	Open Source (Apache-2.0)	rust
Grok Build CLI	AI Lab	both			#7	47%	Subscription	✓	✓	$0 (3 Day Trial) / $30 monthly (SuperGrok) / $300 monthly (SuperGrok Heavy)	✓	✓	✓	✓	✓	✓	—	✓	✓	✓	deep	None - Terminal only	Model-dependent	Open Source (EULA)	rust
Groq Code CLI	Independent	both					BYOK	✓	✓	Free (BYOK)	✓	✗	✓	✗	✗	✗	✗	✗	✓	✓	none	None - Terminal only	Model-dependent	Open Source (MIT)	typescript
IBM Bob	AI Lab	both					Freemium	✓	✗	Free / $20-200/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✗	deep	Bob IDE	Model-dependent	Proprietary Software	javascript
jcode	Independent	both					Freemium, Subscription, BYOK	✓	✓	Free (BYOK) / $10-1,000/mo	✓	✓	✓	—	✓	✓	✓	✓	—	✓	basic	None - terminal only	Model-dependent	Open Source (MIT)	rust
Jules CLI	AI Lab	both					Freemium	✓	✗	Free (15 tasks/day) / $20-125/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✗	deep	None - Terminal only	1M tokens	Proprietary Software	typescript
Junie CLI	Independent	both	#21	71.0%			Freemium	✓	✓	Free (BYOK) / $10-60/mo	✓	✓	✓	✓	✗	✓	✗	✗	✓	✓	deep	JetBrains IDEs	Model-dependent	Proprietary Software	shell
Keen Code	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✗	✗	✗	✗	✓	basic	None - terminal only	Model-dependent; auto-compaction	Open Source (MIT)	go
Kilo Code	Independent	both					Freemium, Usage Based, Subscription, Free, BYOK	✓	✓	Free forever (individual users, BYOK), $19/mo (Kilo Pass), $15/mo (teams)	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	VS Code, JetBrains, Cursor, Windsurf, Trae AI	Model-dependent	Open Source (MIT)	typescript
Kimi Code CLI	Independent	both			#6	52%	Freemium	✓	✓	Free / ~$19/mo	✓	✓	✓	✗	✗	✓	✗	✗	✓	✓	basic	VS Code, Zed, JetBrains	1M (K2.5)	Open Source (Apache-2.0)	python
Kiro CLI	AI Lab	both					Freemium	✓	✗	Free / $0-200/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✗	deep	—	Not disclosed	Proprietary Software	rust
Letta Code	Independent	both	#44	59.1%			Freemium	✓	✓	Free (BYOK) / $20-200/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	basic	None - Terminal only	Model-dependent	Open Source (Apache-2.0)	typescript
Mastra Code	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	—	✓	✓	✓	✓	✓	✓	basic	ACP (Agent Client Protocol)	Model-dependent; observation-based	Open Source (Apache-2.0)	typescript
MiMo Code	AI Lab	both					Freemium, Usage Based, Subscription, BYOK	✓	✓	Free tier / Xiaomi Token Plan CNY 39-659/mo / usage-based API	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	None - terminal only	1M tokens	Open Source (MIT)	typescript
Mistral Vibe	AI Lab	both					Freemium	✓	✓	Free / $15-25/mo	✓	✓	✓	✓	✓	✗	✓	✓	✓	✓	deep	VS Code, JetBrains, Zed	256K tokens	Open Source (Apache-2.0)	python
Muse Code	AI Lab	both			#4	54%	Usage Based	✗	✗	Usage-based (from $0.10/M tokens)	✓	✓	✓	—	✓	✓	✓	✓	—	✗	deep	None - terminal only	1M tokens	Proprietary Software	rust
Nanocode	Independent	both					BYOK	✓	✓	Free (BYOK)	✗	✗	✓	✗	✗	✗	✗	✗	✗	✗	none	None - Terminal only	Model-dependent	Open Source (MIT)	python
Nanocoder	Independent	both					BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✗	✓	✗	✗	✓	✓	✓	basic	VS Code	Model-dependent	Open Source (MIT)	typescript
Neovate Code	Independent	both					BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✗	✓	✗	✓	✗	✓	deep	Zed	Model-dependent	Open Source (MIT)	typescript
omp	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	deep	Zed (via ACP)	Model-dependent; auto-compaction	Open Source (MIT)	typescript
OpenBlock	Independent	both					Freemium	✓	✓	$10-100 (credit-based)	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	None - Terminal only	Model-dependent	Proprietary Software	typescript
OpenCode	Independent	tui	#60	51.7%	#5	54%	Freemium	✓	✓	Free (BYOK) / $10/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	basic	VS Code, Cursor	Model-dependent; auto-compact	Open Source (MIT)	typescript
OpenHands CLI	Independent	cli	#59	51.9%			BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✗	✗	✗	✓	✓	basic	Zed, Toad, VS Code, JetBrains IDEs	Model-dependent	Open Source (MIT)	python
Open Interpreter	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	—	✓	✓	✗	✓	✓	✓	basic	Zed (via ACP)	Model-dependent	Open Source (Apache-2.0)	rust
Pi	Independent	both					BYOK	✓	✓	Free (BYOK)	✗	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	None - Terminal only	Model-dependent; auto-compaction	Open Source (MIT)	typescript
Poolside Agent CLI	AI Lab	tui					Subscription, Subscription, BYOK	✗	✓	$0, Enterprise	✓	✓	✓	✓	✓	✗	✗	✗	✓	✗	deep	Poolside Assistant, VS Code, Visual Studio	(Point) 64K - 128K (Malibu, Laguna)	Proprietary Software	go
Prime Agent	AI Lab	both					Free, Subscription, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	—	✗	✓	✗	✓	✗	✓	basic	Zed, VS Code (via ACP)	Model-dependent	Open Source (MIT)	typescript
Qoder CLI	AI Lab	both					Freemium	✓	✗	Free / $10-200/mo	✓	✓	✓	✓	✓	✓	✗	✓	✓	✗	deep	Qoder IDE, JetBrains	Not disclosed	Proprietary Software	typescript
Qwen Code	AI Lab	both					Free	✓	✓	Free (1,000 req/day)	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	basic	VS Code, Zed, JetBrains IDEs	Model-dependent	Open Source (Apache-2.0)	typescript
Ralph TUI	Independent	tui					BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✗	✓	✓	✗	✗	✓	✓	deep integration with git worktrees	None - Terminal only	Depends on underlying agent	Open Source (MIT)	typescript
Rovo Dev CLI	Independent	both					Subscription	✓	✗	$20/mo per user	✓	✓	✓	✓	✓	✓	✗	✓	✗	✗	deep	VS Code, Bitbucket Cloud	Model-dependent	Proprietary Software	python
Snowflake Cortex CLI	Independent	both					Freemium	✓	✗	Free trial / $20/mo	✓	✓	✓	✗	✓	✗	✗	✓	✓	✗	deep	None - Terminal only	Model-dependent	Proprietary Software	typescript
Standard Code	Independent	both					Subscription, BYOK	✗	✓	$49/mo per line ($5 first week)	✓	✓	✓	✓	✓	✓	—	✓	✓	—	basic	None - terminal only	—	Proprietary Software	typescript
Toad	Independent	both					Free	✓	✓	Free	✓	✓	✓	✗	✗	✓	✗	✓	✓	✓	deep	None - Terminal only	Depends on underlying agent	Open Source (AGPL-3.0)	python
Vix	Independent	tui					Free	✗	✓	$0	✓	✓	✗	✗	✓	✓	✗	✓	✓	✗	shallow	N/A	Model-dependent	Open Source (AGPL-3.0)	go
VT Code	Independent	both					BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✗	✓	✓	✗	✓	✓	✓	basic	VS Code, Zed, Cursor, Windsurf	Model-dependent; auto-summarization	Open Source (MIT)	rust
Warp Agent CLI	Independent	cli	#40	61.2%			Freemium, Subscription, Usage Based, BYOK	✓	✓	Free / $20-200/mo	✓	✓	✓	—	✓	✓	✓	—	✓	✓	basic	None - terminal only	Model-dependent	Proprietary Software	rust
Zero	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✓	✓	✓	✗	✓	deep	Zed, JetBrains, Neovim (via ACP)	Model-dependent; auto-compaction	Open Source (MIT)	go
zerostack	Independent	both					Free, BYOK	✓	✓	Free (BYOK)	✓	✓	✓	✓	✓	✓	✗	✓	✓	✓	deep	Zed (via ACP), Neovim (via CodeCompanion.nvim + ACP)	Model-dependent; auto-compaction	Open Source (GPL-3.0)	rust
Zot	Independent	both					Free	✗	✓	$0	✗	✓	✓	✗	✓	✓	✗	✓	✓	✓	low	None	Model dependent	Open Source (MIT)	go
Continue	Independent	both					Freemium	✓	✓	Free (BYOK) / $10/mo	✓	✗	✓	✓	✓	✗	✗	✓	✓	✓	basic	VS Code, JetBrains IDEs	Model-dependent	Open Source (Apache-2.0)	typescript
Plandex	Independent	cli					Free	✓	✓	Free (self-hosted)	✗	✗	✗	✓	✓	✗	✗	✗	✗	✓	deep	None - Terminal only	2M tokens effective	Open Source (MIT)	go

Last updated September 2026.

TBench scores from Terminal Bench. Each agent shows its most recent benchmark generation, so scores in this column may come from different versions and are not directly comparable. AA Rank and AA Score from the Artificial Analysis Coding Agent Index, a 0 to 100 composite for an agent paired with a specific model. The two columns use different scales and are not comparable with each other.

0 agents selected

Showing 65 of 65 agents

Provider:

Source:

Pricing:

AI Coding Agents Overview

As of September 2026, Terminal Trove tracks 65 terminal AI coding agents, also called AI coding assistants. 63 are actively developed and 42 are open source.

60 support MCP, 56 offer a free tier, 51 support bring your own key, and 49 can run local models.

15 carry Terminal Bench results across 4 boards. On Terminal-Bench 4.0, the newest, Codex CLI logo Codex CLI leads at 58.2%, ahead of Claude Code logo Claude Code at 57.9%.

Terminal-Bench 2.0 has the widest coverage at 12 agents. Codex CLI logo Codex CLI leads there on 82.2%, which places it #4 on the public leaderboard. Accuracy falls as the boards get harder, so scores are not comparable between versions.

8 appear on the Artificial Analysis Coding Agent Index, a separate composite score that rates an agent together with the model it runs. The highest pairing is Claude Code logo Claude Code on Fable 5.1 (max, with fallback) at 62%.

Feature Guide

A comprehensive reference for each AI coding agent feature tracked by Terminal Trove. Click any feature name for a detailed explanation, adoption stats, and which agents support it.

Jump to Feature

Code Intelligence

Multi-Agent Support· Multi-File Editing

Integration & Connectivity

MCP Support· Hooks / Extensions· Web Search· Local Model Support

Automation & Security

Auto Test Running· Sandbox / Permissions

Input & Context

Image / Multimodal Input· Voice Mode

Code Intelligence

Multi-Agent Support

Supports running multiple AI agents concurrently or orchestrating sub-agents to divide complex tasks into parallel workstreams.

Agent Support 59 of 65 (90%)

Multi-File Editing

Can read, create, and modify multiple files in a single operation. Essential for refactoring, adding features that span several modules, or generating boilerplate across a project.

Agent Support 63 of 65 (96%)

Integration & Connectivity

MCP Support

Supports the Model Context Protocol (MCP), an open standard for connecting AI agents to external tools, data sources, databases, and APIs through a unified interface.

Agent Support 60 of 65 (92%)

Hooks / Extensions

Supports lifecycle hooks or a plugin/extension system that lets you run custom scripts or commands at specific points during the agent's workflow.

Agent Support 53 of 65 (81%)

Web Search

Can search the web to find documentation, examples, and up-to-date information while working on your code. Helps the agent stay current beyond its training data.

Agent Support 55 of 65 (84%)

Local Model Support

Can run with locally hosted AI models (e.g. Ollama, llama.cpp) instead of cloud APIs. Enables fully offline, private development without sending code to external servers.

Agent Support 49 of 65 (75%)

Automation & Security

Auto Test Running

Automatically runs your test suite after making code changes and iterates on failures. The agent can detect test failures and fix them without manual intervention.

Agent Support 45 of 65 (69%)

Sandbox / Permissions

Runs commands in a sandboxed environment with configurable permissions. Prevents the agent from executing dangerous operations without explicit approval.

Agent Support 56 of 65 (86%)

Input & Context

Image / Multimodal Input

Accepts images, screenshots, and other visual input alongside text. Useful for UI work, debugging visual issues, or providing design mockups as context.

Agent Support 54 of 65 (83%)

Voice Mode

Supports voice input for dictating instructions or having a spoken conversation with the agent. Enables hands-free coding workflows.

Agent Support 19 of 65 (29%)

Aider

Auggie CLI Blackbox AI CLI logo

Blackbox AI CLI Claude Code logo

Claude Code Cline CLI logo

Cline CLI

Codewhale

Codex CLI

Cursor CLI Gemini CLI logo

Gemini CLI GitHub Copilot CLI logo

GitHub Copilot CLI Goose CLI logo

Goose CLI

jcode

Mastra Code MiMo Code logo

MiMo Code

Mistral Vibe Muse Code logo

Muse Code

omp

Warp Agent CLI Zero logo

Zero

Frequently Asked Questions

What is the best free AI coding agent?

There are 59 AI coding agents with free tiers or free pricing. Among them, Droid CLI logo

Droid CLI has the highest Terminal Bench accuracy at 77.3%.

Which AI coding agents support MCP?

60 AI coding agents support MCP (Model Context Protocol):

What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard that allows AI coding agents to connect to external tools, data sources, and services.

It enables agents to access files, run commands, query databases, and interact with APIs through a unified interface.

MCP support means an agent can be extended with third-party integrations.

What is Terminal Bench?

Terminal Bench is an open-source benchmark for evaluating AI coding agents on real-world terminal tasks.

It measures accuracy across standardized scenarios.

Learn more at tbench.ai.

What is the Artificial Analysis Coding Agent Index?

The Artificial Analysis Coding Agent Index is a composite score from 0 to 100, built from three benchmarks: DeepSWE v1.1, Terminal-Bench 4.0 and SWE-Atlas-QnA, equally weighted. Higher is better.

It measures an agent paired with a specific model, not the agent alone, so the same agent scores differently depending on what it runs.

It is not the same number as the Terminal Bench columns, even though Terminal-Bench 4.0 is one of its three parts. Artificial Analysis runs its own evaluations rather than reusing the tbench.ai leaderboard, and averages that result with two other benchmarks, so the two columns disagree on the same pairing and are not comparable.

Learn more at artificialanalysis.ai.

Which AI coding agents are open source?

42 AI coding agents are open source:

Which AI coding agents support BYOK (Bring Your Own Key)?

51 AI coding agents support BYOK, letting you use your own API keys from providers like OpenAI, Anthropic, or Google:

How do AI coding agents compare on pricing?

Among 65 tracked agents: 17 are Free, 26 are Freemium, 30 are BYOK, 15 are Subscription, and 10 are Usage Based.

Which AI coding agent has the best benchmark score?

The top-ranked AI coding agent on Terminal-Bench 2.0 is Codex CLI logo

Codex CLI at #4 with 82.2% accuracy.

On the separate Artificial Analysis Coding Agent Index, the highest scoring pairing is Claude Code logo Claude Code on Fable 5.1 (max, with fallback) at 62%.

What is the best AI coding assistant?

The best AI code assistant depends on your workflow and budget.

For terminal-based development, agents like Claude Code logo Claude Code, Aider logo Aider, and Cursor CLI logo Cursor CLI lead in Terminal Bench accuracy.

Use the comparison table above to filter by features like MCP support, BYOK, multi-file editing, and pricing to find the best fit.

Request Information Update