Updated June 12, 2026 · Verified Sources

CLI AI Coding Agents —
The 2026 Deep Comparison

From Anthropic's Claude Opus 4.8 and Google's new Antigravity CLI to China's Qwen 3.6, Kimi K2.6 and xAI's Grok Build — every major terminal agent compared on price, context, model, MCP support & more.

12
Agents Covered
20+
Criteria
3
Chinese Agents
$0
Min. to Start

🤖 Agent Profiles

Every major CLI coding agent as of June 12, 2026 — including 2026 model bumps, pricing changes & new entrants.

🤖
Claude Code
Anthropic
Pro $20/mo Open Source MCP

Anthropic's flagship terminal agent. Now defaults to Claude Opus 4.8 (May 28, 2026) with dynamic workflows for large-scale tasks. June 15 billing restructure splits programmatic usage onto a separate credit pool. Safe Mode & plugin marketplace shipped in 2026.

$ npm install -g @anthropic-ai/claude-code
Key Specs
Context1M tokens
ModelOpus 4.8 (default)
Free Tier$5 credit
Paid$20–$200/mo
Capabilities
✦ Sub-agents, parallel tasks
✦ Plugin marketplace (2026)
✦ Safe Mode for diagnostics
✦ Doubled rate limits (May 5)
✦ CLAUDE.md + hooks + MCP
◈ Jun 15: API-rate credits for SDK
51.9%
SWE-bench Pro
1M
Context
9.4
Ecosystem
⚡
Codex CLI
OpenAI
Plus $20/mo Open Source MCP

OpenAI's open-source terminal agent. Now running GPT-5.5 (Apr 23) and GPT-5.3-Codex (Feb 5, Terminal-Bench 2.0 SOTA). Migrated to Rust (codex-rs, 95.6% of repo). /goal Goal Mode GA in v0.133 (May 21).

$ npm install -g @openai/codex
Key Specs
Context~400K
ModelGPT-5.5 / 5.3-Codex
Free TierNone
Paid$20–$200/mo
Capabilities
✦ /goal Goal Mode GA (v0.133)
✦ Rust rewrite (codex-rs)
✦ Chrome DevTools / CDP debug
✦ Enterprise sandbox + Python SDK
✦ AGENTS.md + MCP + multimodal
◈ No permanent free tier
59.1%
SWE-bench Pro
400K
Context
9.1
Ecosystem
🌟
Antigravity CLI
Google (ex-Gemini CLI)
Freemium Closed-source (2026) MCP

Announced at I/O 2026 (May 19) as the replacement for Gemini CLI — which is deprecated June 18, 2026 for free/individual users. Go rewrite, now closed-source (was Apache 2.0); community backlash after 6,000+ PRs locked out. Ships with Gemini 3.5 Flash.

$ curl -fsSL antigravity.dev/install | bash
Key Specs
Context1M tokens
ModelGemini 3.5 Flash / 3.1 Pro
Free TierFlash-only (Mar 25+)
PaidAI Ultra $100/mo
Capabilities
✦ Gemini 3.5 Flash (4× faster)
✦ Skills, Hooks, Subagents
✦ Antigravity plugins (ex-Extensions)
✦ Google Search grounding + MCP
✕ Closed-source (backlash)
◈ Free 1000 req/day ended
43.3%
SWE-bench Pro
1M
Context
7.6
Ecosystem
🐙
Copilot CLI
GitHub (Microsoft)
Pro $10/mo MCP GitHub Native

Reached GA on Feb 25, 2026. Default model upgraded to Claude Sonnet 4.6 (Feb 17; Sonnet 4 deprecated May 6). /fleet parallel multi-agent shipped Apr 1. Microsoft's in-house Project Polaris replaces GPT-4 Turbo in August 2026.

$ npm install -g @github/copilot
Key Specs
Context200K+
ModelSonnet 4.6 (Polaris Aug)
Free TierLimited
Paid$10–$39/mo
Capabilities
✦ GA: Plan + Autopilot modes
✦ /fleet parallel multi-agent
✦ Persistent .copilot/memory/
✦ Custom agents (.github/extensions)
✦ Background & delegation
◈ Paid Copilot plan required
GA
Feb 25, 2026
200K
Context
9.2
Ecosystem
☁️
Amazon Q CLI
AWS (Amazon)
Free Tier AWS Native MCP

AWS's developer CLI, deeply integrated with the AWS ecosystem. Free tier: 50 chat interactions/month. Now powered by Claude Sonnet 4. AWS is increasingly positioning Kiro as its agentic CLI flagship, while Q remains the AWS-infra workhorse.

$ brew install amazon-q
Key Specs
Context100K chars
ModelClaude Sonnet 4
Free Tier50 msg/mo
PaidPro $19/user/mo
Capabilities
✦ AWS resource management
✦ CloudFormation & IaC review
✦ MCP server support
✦ Java 8→17 & .NET migration
✦ Resumable conversations
◈ Very limited free (50 msg/mo)
~58%
SWE-bench
100K
Context
9.8
AWS Depth
🎯
Kiro CLI
Amazon Web Services
Free 50 cr/mo Spec-Driven · GA MCP

Now GA (re:Invent Nov 2025) with permanent freemium tier. New Kiro Pro Max $100/mo tier (May 2026). May 12 shipped Parallel Task Execution + Requirements Analysis engine. Sonnet 4.6 / Opus 4.8 on paid plans. GovCloud + GitLab + TDD mode in 2026.

$ npm install -g @kirodotdev/kiro
Key Specs
Context200K+
ModelSonnet 4.6 / Opus 4.8
Free Tier50 credits/mo
Paid$20–$200/mo
Capabilities
✦ Requirements Analysis engine
✦ Parallel Task Execution (May 12)
✦ TDD mode + Quick Plan workflow
✦ GovCloud / FedRAMP High
✦ GitLab + GitHub integration
◈ Credit-based (overage $0.04/cr)
9.5
Spec Quality
$100
Pro Max
9.4
Innovation
🔥
Qwen Code
Alibaba Cloud (China)
API paid · BYOK Apache 2.0 Chinese

Alibaba's CLI agent — Gemini CLI fork now on Qwen 3.6 Plus (Apr) with 1M context, 78.8% SWE-bench Verified. Free 2,000 req/day tier ended Apr 15, 2026; switch to Bailian paid API or self-host. v0.18.0 added ACP mode + /skills.

$ npm install -g @qwen-code/qwen-code
Key Specs
Context1M tokens
ModelQwen 3.6 Plus / 3.7 Max
Free TierEnded Apr 15
LicenseApache 2.0 (CLI)
Capabilities
✦ Qwen 3.6 Plus: 78.8% SWE-bench
✦ ACP mode + /skills picker (v0.18)
✦ Any OpenAI-compatible API (BYOK)
✦ Open-weight 3.6-35B (Apache 2.0)
◈ Free 2K req/day ended Apr 15
◈ ~$0.28/M input (Bailian)
78.8%
SWE-bench Verified
1M
Context
12×
Cheaper vs Opus
🌙
Kimi Code CLI
Moonshot AI (China)
Free via plan Open Source Chinese

Moonshot's open-source CLI agent. Kimi K2.6 reached GA Apr 21, 2026 — 1T MoE / 32B active, native multimodal (MoonViT), Agent Swarm of up to 300 sub-agents coordinating 4,000+ steps. Modified MIT license.

$ pip install kimi-code && kimi
Key Specs
Context256K
ModelKimi K2.6 (GA Apr 21)
Free TierVia $20 plan
SWE-bench Pro58.6%
Capabilities
✦ 300-subagent Swarm / 4K steps
✦ Native multimodal (MoonViT)
✦ ACP (Zed, JetBrains) + MCP
✦ HLE-Full 54.0% (claimed lead)
✦ Open weights (Modified MIT)
◈ Weekly quota caps on plans
58.6%
SWE-bench Pro
300
Sub-agents
9.4
Value
⚙️
Trae Agent
ByteDance (China)
Free (BYOK) Open Source MIT MCP (2026) Chinese

ByteDance's research agent + commercial TRAE CLI. MCP support added in 2026 (stdio, SSE, Streamable HTTP). TRAE SOLO rebranded to TRAE Work in June. Trajectory recording for audit. MIT license.

$ git clone github.com/bytedance/trae-agent
Key Specs
ContextDepends on API
ModelBYOK (any)
Free TierYes (BYOK)
LicenseMIT
Capabilities
✦ Fully local — no third-party
✦ Trajectory recording (audit log)
✦ MCP: stdio + SSE + HTTP (new)
✦ Anthropic + OpenAI APIs
✦ MIT — fully forkable
◈ No built-in free quota
~55%
SWE-bench
9.8
Transparency
9.0
Flexibility
🖥️
Warp Code
Warp
Freemium AGPL-3.0 (May 2026) MCP

AI-native terminal — open-sourced May 7, 2026 under AGPL-3.0 (UI crates MIT) with OpenAI as founding sponsor. May 2026 release added Codex as harness for local child agents, drag-drop images, inline Mermaid + image rendering, and configurable max context. Cloud orchestration platform (Oz) remains proprietary.

$ brew install --cask warp
Key Specs
ContextConfigurable max
ModelClaude / GPT-5.5 + Codex harness
Free TierFreemium + BYOK
Paid$0.04/credit overage
Capabilities
✦ Open-sourced (AGPL-3.0, 56K★)
✦ Codex as local child-agent harness
✦ Drag-drop images + Mermaid render
✦ Cited #1 Terminal-Bench (vendor)
◈ Oz cloud platform remains proprietary
9.4
UX
9.2
Git Workflow
8.4
Flexibility
🛸
Grok Build
xAI
Beta (May 25, 2026) ACP Parallel Sub-Agents

xAI's terminal coding agent launched May 25, 2026 for SuperGrok + X Premium Plus subscribers. Plan mode, 8 parallel sub-agents in isolated git worktrees, headless CI mode, and Agent Client Protocol (ACP) support.

$ curl -fsSL https://x.ai/cli/install.sh | bash
Key Specs
Context2M tokens
Modelgrok-build-0.1
Free TierVia SuperGrok / X Premium+
Paid$30/mo (SuperGrok)
Capabilities
✦ Plan mode before execution
✦ 8 parallel sub-agents (git worktrees)
✦ Headless CI mode
✦ Agent Client Protocol (ACP)
◈ Beta — feature set still evolving
2M
Context
8×
Parallel Agents
Beta
Status
🧩
OpenCode
opencode.ai
Open Source Free (Local) 75+ Providers

Open-source multi-model terminal UI that connects to 75+ providers — Anthropic, OpenAI, Gemini, Qwen, Kimi, and any local model via Ollama or LM Studio. Completely free when run against local models.

$ npm install -g opencode-ai
Key Specs
ContextModel-dependent
ModelAny (75+ providers)
Free TierYes (local models)
PaidBYOK only
Capabilities
✦ 75+ provider integrations
✦ Ollama / LM Studio support
✦ Polished terminal UI
✦ Fully open-source
◈ Quality depends on chosen model
75+
Providers
$0
Local Cost
OSS
License

📊 Side-by-Side Comparison

All 12 agents across key criteria — updated June 2026.

Agent SWE-bench Context Free Tier Min. Price MCP Open Source Best For
🤖 Claude CodeAnthropic51.9% Pro1M$5 credit$20/mo✔✔Best overall (Opus 4.8)
⚡ Codex CLIOpenAI59.1% Pro400K✗$20/mo✔✔GPT-5.5 ecosystem
🔥 Qwen CodeAlibaba78.8% Ver.1MEnded Apr 15BYOK✗✔BYOK / self-host
🌙 Kimi Code CLIMoonshot AI58.6% Pro300K sub-agentsVia plan$20/mo✔✔Value + multimodal
🌟 Antigravity CLIGoogle43.3% Pro1M1000/day$0 (EOL Jun 18)✔ClosedDeprecated soon
🐙 Copilot CLIGitHub~70%200KLimited$10/mo✔✗GitHub workflows
🎯 Kiro CLIAWS—200K+Preview$100/mo Pro Max✔✗Spec-driven dev
☁️ Amazon Q CLIAWS~58%100K50 msg/mo$0✔✗AWS (→ Kiro)
⚙️ Trae AgentByteDance~55%API-dep.BYOK$0MCP 2026✔Privacy / audit
🖥️ Warp CodeWarp—ConfigurableFreemium$0✔AGPL-3.0Open terminal UX
🛸 Grok BuildxAI—2MVia SuperGrok$30/moACP✗Parallel sub-agents
🧩 OpenCodeopencode.ai—Model-dep.Free (local)$0✔✔Multi-provider TUI

🧭 Decision Guide

Pick the right agent for your use case.

🏆 Best Overall

Need the highest quality and widest ecosystem? Nothing beats Claude Code for autonomous tasks and SWE-bench scores.

→ Claude Code

🆓 Best Free Option

Want a free powerful agent with no credit card? Antigravity CLI gives you 1M context and 1000 requests/day at $0.

→ Antigravity CLI

💰 Best Value

Want near-Claude-Code performance for $20/mo with generous weekly quota? Kimi Code CLI at 76.8% SWE-bench is the value pick.

→ Kimi Code CLI

☁️ AWS / Cloud Infra

Working heavily with AWS, EC2, S3, IAM, or CloudFormation? Amazon Q CLI is purpose-built for this — nothing else comes close.

→ Amazon Q CLI

🐙 GitHub-First Teams

Using GitHub Issues, PRs, and Actions daily? Copilot CLI integrates natively and works on your existing $10/mo plan.

→ Copilot CLI

🔒 Privacy / Audit

Need a fully local agent with no vendor routing and full trajectory logs for audit compliance? Trae Agent is MIT-licensed and BYOK.

→ Trae Agent

📋 Spec-Driven Teams

Want requirements → design → tasks before a single line of code is written? Kiro CLI's neurosymbolic spec approach is unique in this space.

→ Kiro CLI

🇨🇳 Self-Hostable BYOK

Want to use any OpenAI-compatible API with 2000 free runs/day and Apache 2.0 licensing? Qwen Code lets you bring your own key.

→ Qwen Code

📚 Sources

All data verified from official documentation and benchmarks.

2OpenAI Codex CLI — github.com/openai/codex
3Antigravity CLI (Google) — github.com/google-gemini/gemini-cli
4GitHub Copilot CLI — docs.github.com/en/copilot
5Amazon Q Developer — aws.amazon.com/q/developer
6Kiro CLI — kiro.dev
7Qwen Code (Alibaba) — github.com/QwenLM/qwen-code
8Kimi Code CLI (Moonshot) — github.com/MoonshotAI/kimi-cli
9Trae Agent (ByteDance) — github.com/bytedance/trae-agent
10Warp Code (open-sourced May 2026) — dev.to coverage
11Grok Build CLI (xAI) — launch review
12OpenCode — opencode.ai/docs