Blog

Claude Code alternatives in November 2026: 10 coding agents compared on cost, models and open source

Ramashankar S · · Updated · 19 min read

The main Claude Code alternatives in November 2026 are OpenAI's Codex CLI, Google's Antigravity CLI, OpenCode, Cline, Kilo Code, Cursor, GitHub Copilot, Aider, Amp and Replit Agent. Codex CLI is the closest swap: a terminal agent included in the $20 ChatGPT Plus plan. OpenCode, Cline, Kilo Code and Aider are free, open source and run any model on your own API keys, so the model sets the bill: the same month of daily agent work costs $7.47 on DeepSeek V4.1 Flash off-peak and $282.24 on Opus 5.5.

This post compares tools; for which model to run, see Claude alternatives for coding. Everything below was checked on 5 November 2026.

Claude Code alternatives at a glance

Agent (version checked)Where it runsOpen sourceModelsPaid plans from
Claude Code 2.1.318 (baseline)Terminal, IDEs, desktop, webNoClaude only$20 (Claude Pro)
Codex CLI 0.166.0Terminal, IDEs, desktop, webCLI yes (Apache-2.0)OpenAI; Responses API endpoints, Ollama, LM Studio$20 (ChatGPT Plus)
Antigravity CLITerminal, plus app and IDE extensionsNoGemini, plus Claude on paid plans; no own keysFree; $19.99 (Google AI Pro)
OpenCode 2.2.5Terminal, desktop, webYes (MIT)Any provider, own keysFree; $10 (OpenCode Go)
Cline 4.1.26VS Code, CLI, desktopYes (Apache-2.0)Any provider, own keysFree; $9.99 (ClinePass)
Kilo Code 7.8.25VS Code, JetBrains, CLI, cloudYes (MIT)500+ through Kilo Gateway, or own keysFree; $19 (Kilo Pass)
Cursor 3.28.7Its own editor, CLI, cloudNoGrok and Composer in the plan; Claude and Gemini at API list price$8 (Plus, Grok only)
GitHub Copilot, CLI 1.0.105IDEs, CLI, cloud agent on GitHubNoClaude, GPT, Gemini, Grok, Kimi; own keys in the CLI$10 (Pro)
Aider 0.86.2TerminalYes (Apache-2.0)Any provider, own keysFree
AmpCLI, web, mobile apps, cloud machinesNoAmp's picks, or own keys and subscriptionsFree; $20 (Megawatt)
Replit AgentBrowser, with hostingNoAutomatic in Free Mode; Claude, GPT, Kimi in paid modesFree; $20 (Core)

Older lists miss three 2026 changes. On 18 June Google moved free and Google AI Pro or Ultra users from Gemini CLI to Antigravity CLI. Roo Code's repository was archived after a last release on 15 May, and roocode.com redirects to Roomote. And Cursor, bought by SpaceX in August, sells new plans that include only Grok and Composer models.

What Claude Code gives you, and what it locks in

Claude Code comes with every paid Claude plan: Pro at $20 a month ($200 a year), Max at $100 and Ultra at $200, with limits covered in the Claude Code pricing guide. Its default model is Opus 5.5 on every plan and on the API.

The lock-in is the model. Anthropic says it doesn't support routing Claude Code to non-Claude models through any gateway, and the repository is "All rights reserved". People look elsewhere to run GPT, Gemini or open-weight models, to pay less when plan limits run out, to get code they can audit, or to work in an editor or browser.

What a month of agent work costs

The chart prices one workload on each route: 20 days of 300 model calls, 60,000 tokens of context per call (92% cache reads, 8% cache writes) and 600 output tokens per call, the "daily" profile from the pricing guide. Per-token rows use each maker's list price in a bring-your-own-key (BYOK) agent such as OpenCode, Cline, Kilo Code or Aider.

Monthly cost of the same daily agent workload on 13 routes, from $7.47 to $282.24
RoutePer monthWhat it covers
Any BYOK agent + DeepSeek V4.1 Flash$7.47 off-peak, $14.95 peakDeepSeek's list price
OpenCode Go$10.00Open-weight models, monthly allowance per model
Antigravity CLI on Google AI Pro$19.99Limits not published in tokens
Codex CLI on ChatGPT Plus$20.0015 to 160 GPT-6.1 Sol messages per 5 hours
Claude Code on Pro / Max$20.00 / $100.005-hour and weekly limits
GitHub Copilot Max, Sonnet 5.5$100.00Inside $200 of included usage
GitHub Copilot Pro+, Sonnet 5.5$110.12$39 plus usage past $70
Any BYOK agent + Sonnet 5.5 or GPT-6.1 Sol$141.12Both $2 / $10 per 1M tokens
Cursor Premium + Sonnet 5.5$171.12$30 plus Claude at list price
GitHub Copilot Max, Opus 5.5$182.24$100 plus usage past $200
Opus 5.5 per token, any agent$282.24Claude Code's default model on an API key

For Claude models, a Claude plan is the lowest-cost route while its limits hold: Max at $100 against $141.12 for Sonnet 5.5 or $282.24 for Opus 5.5 per token. Only Copilot Max matches it, because its $200 of included usage covers the Sonnet month. The real savings come from changing the model: DeepSeek V4.1 Flash costs about 5% of the Sonnet month off-peak, and its off-peak window covers the US working day. No vendor publishes plan limits in tokens, so check a week of real usage before relying on a flat price.

How the agents score

The agent around a model changes results. Artificial Analysis runs agent and model pairs through FrontierCode v1.1, Terminal-Bench 4.0 and SWE-Atlas-QnA, three attempts per task, and reports cost per task at API prices.

Artificial Analysis Coding Agent Index scores, cost per task and time per task for nine agent and model pairs
AgentModelIndexCost per taskTime per task
Claude CodeOpus 5.5, max61.7$11.4261 min
Devin Fusion CLIOpus 5.5 + GPT-6 Luna61.5$8.0051 min
GitHub CopilotCopilot Fusion, max59.9$5.5936 min
Claude CodeSonnet 5.5, max59.3$13.3080 min
CodexGPT-6.1 Sol, max54.1$1.5024 min
Antigravity CLIGemini 4 Argon, high53.2$11.7335 min
OpenCodeGLM-5.3, max48.9$5.2356 min
CodexDeepSeek V4.1 Flash, max47.3$0.3938 min
Grok BuildGrok 4.7, xhigh46.6$9.4836 min

Claude Code with Opus 5.5 still leads. Copilot's Fusion router, which picks between Claude and GPT models, comes within two points at under half the cost per task, and Codex with GPT-6.1 Sol scores 54.1 for $1.50. The official Terminal-Bench 4.0 leaderboard agrees: 64.85% for Opus 5.5 in Claude Code, 58.18% for GPT-6.1 Sol in Codex. Neither board has run Cline, Kilo Code, Cursor, Aider, Amp or Replit Agent.

The alternatives one by one

Codex CLI

The closest like-for-like replacement: a terminal agent with an IDE extension, a desktop app and cloud tasks. It comes with ChatGPT Plus at $20 (15 to 160 GPT-6.1 Sol messages per five hours, plus weekly limits) and Pro at $100, $200 or $500, with no five-hour limit, or bills API prices on an API key.

The CLI is Apache-2.0; the IDE extension and cloud service aren't open source. Built-in providers include Ollama, LM Studio, Amazon Bedrock and Google Cloud, and codex --oss runs a local model, but custom providers must speak the Responses API: current releases don't support Chat Completions-only endpoints. See Codex vs Claude Code.

Antigravity CLI (Gemini CLI's replacement)

Gemini CLI 0.66.0 is still Apache-2.0, but its README limits it to Gemini Code Assist Standard and Enterprise licenses and paid API keys. Individuals now get Antigravity CLI, run as agy, which reads GEMINI.md and AGENTS.md and imports Gemini extensions with agy plugin import gemini.

The free tier has a weekly quota on Gemini 3.8 Flash. Google AI Pro at $19.99 adds a five-hour quota, Gemini 4 Argon, and Claude Sonnet 5.5 and Opus 5.5 on paid (not trial) accounts; Ultra costs $99.99 or $199.99. Google lists 18 November as the date Claude Sonnet 4.6 and Opus 4.6 leave the free tier. There's no bring-your-own-key option, and the CLI ships as a closed binary. See Gemini CLI vs Claude Code.

OpenCode

The most-starred agent in this comparison (about 224,000 GitHub stars, MIT). Version 2 runs in the terminal, a desktop app and a browser, and connects to any provider in the models.dev catalog, a ChatGPT Plus or Pro sign-in, or any OpenAI-compatible endpoint.

OpenCode Go, $10 a month ($40 for Go Plus), covers open-weight models with a monthly allowance per model, such as $100 of DeepSeek V4.1 Flash usage. Console pay-as-you-go, formerly OpenCode Zen, charges list price plus 4.6% + $0.31 per purchase. On 17 October OpenCode reported a breach affecting 1.4% of Console users who had linked outside provider keys; the coding agent wasn't affected. More in OpenCode vs Claude Code.

Cline

A VS Code extension, CLI and desktop app, free and Apache-2.0, that asks for approval before file edits and commands by default. It works with 30+ providers, any OpenAI-compatible endpoint and local models through Ollama or LM Studio. ClinePass, $9.99 a month, covers open models such as DeepSeek V4.1 Flash, GLM 5.3 and Kimi K3 at 2 to 5 times standard rate limits. Its Claude Code provider runs the claude CLI on your Pro or Max plan. JetBrains support is in the Enterprise tier. See what Cline is and how it works.

Kilo Code

MIT-licensed, with extensions for VS Code and JetBrains, a CLI and cloud agents from $0.60 an hour. Kilo Gateway sells 500+ models at the provider's rate with no markup, plus a 5% fee when you pay; own keys and local models cost nothing extra. Kilo Pass, from $19 a month, adds extra usage on top (up to $26.60 a month on the $19 tier). Teams cost $15 per user. Kilo publishes a migration guide for Roo Code users.

Cursor

An editor first, with autocomplete, inline edits, an agent, a CLI and cloud agents. Since 12 October it's sold as The X Subscription: Plus at $8 runs only Grok and Composer, and Premium ($30), Super ($100) and Ultra ($200) add third-party models such as Claude at API list price, billed separately. OpenAI set a 12 November shutoff for its models in Cursor. You can still run the Claude Code extension inside Cursor on a Claude plan. See Cursor vs Claude Code.

GitHub Copilot

Since 1 June, Copilot bills by token at list prices, Claude at Anthropic's own rates, with included usage of $15 on Pro ($10), $70 on Pro+ ($39) and $200 on Max ($100), per GitHub's plans page. The model menu is the widest here: Claude up to Opus 5.5 (Pro+ and above), GPT-6.1 Sol, Gemini 4 Argon, Grok 4.7 and Kimi K3, plus the Fusion router. Every plan includes Copilot CLI, whose providers.json accepts your own OpenAI-compatible, Azure or Anthropic endpoint. The cloud agent takes GitHub issues and opens pull requests. See Claude Code vs GitHub Copilot.

Aider

A terminal pair programmer, Apache-2.0. You add files to the chat, it edits them and commits each change to git. It connects to any OpenAI-compatible API with OPENAI_API_BASE and --model openai/<model>, and returns edits as search-and-replace text blocks, so it doesn't need tool calling. The catch is pace: the latest release, 0.86.2, came out on 12 February 2026, and the last commit was on 22 May.

Amp

Amp runs agents on cloud machines it calls orbs, which you start and check from the CLI, the web or phone apps. Pricing: Hobby is free with nothing included, Megawatt is $20 with $20 of usage and 750 orb hours, Gigawatt is $200 with $200 of usage. Amp adds no markup for individuals. Its model routing takes your own API keys, a ChatGPT or Copilot subscription, or any OpenAI- or Anthropic-compatible URL, and a Claude Code mode runs Anthropic's Agent SDK on a Claude plan.

Others worth knowing

Windsurf's pricing page now redirects to Devin, at $20 or $200 a month; its Fusion CLI scored 61.5 above. Grok Build, xAI's agent, comes with Cursor's X Subscription plans. Smaller terminal agents include Qwen Code (0.26.7, Apache-2.0), Goose (Apache-2.0) and Charm's Crush (0.99.0, source-available under FSL-1.1-MIT).

OpenCode 2.0: install, free and local models, background agentsLeon van Zyl · 14:55 · YouTube
A walkthrough of OpenCode 2.0, published 22 October 2026, building a small app with each new feature.

Replit vs Claude Code

Replit Agent builds and hosts an app for you in the browser; Claude Code works on a codebase on your machine. Pick Replit for a running app with a URL and no local setup, and Claude Code when you have a repository, tests and your own deployment.

Replit AgentClaude Code
Where it runsReplit's cloud workspace, in a browserYour terminal, IDE or desktop app
What you getApp, database, hosting and a public URLEdits, commits and pull requests in your repo
ModelsAutomatic in Free Mode; pick Claude, GPT or Kimi in Power and MaxClaude only, Opus 5.5 by default
PriceStarter free; Core $20 ($18 a month yearly); Pro $100Pro $20; Max $100; Ultra $200
How usage is billedFree Mode allowance per five hours and week, then effort-based charges per task5-hour and weekly limits, then API rates if you opt in
Existing codebaseImports GitHub or GitLab reposBuilt for it

Replit's Core plan includes $20 of usage and up to 30 hours of chat in Free Mode, and Pro includes $100 that rolls over for two months. Power and Max modes charge per finished request by effort, and outside model calls bill at the provider's public rate, per the AI billing docs. That's harder to predict than a flat Claude plan, so set a spend limit.

Replit's CEO said on 16 October that OpenAI is now Replit Agent's default, with Claude still selectable, TechCrunch reported. Replit's model-selector page and changelog list slightly different models, so trust the selector in your account.

Pick by situation

Decision guide: which Claude Code alternative fits ten common situations, with monthly prices
  • For Claude models and an agent most days, stay on Claude Code, Pro or Max ($20 to $100).
  • For OpenAI models or a second reviewer, add Codex CLI on ChatGPT Plus ($20).
  • To start with no spend, use Antigravity CLI's free tier or OpenCode's free models.
  • For open-weight models at a flat price, take OpenCode Go ($10) or ClinePass ($9.99).
  • For any model on your own keys in VS Code or JetBrains, use Cline or Kilo Code.
  • If work starts from GitHub issues, use Copilot Pro+ or Max ($39 to $100).
  • If you type most code yourself, use Cursor Premium ($30 plus usage).
  • For a small git-first terminal tool that works with any API, use Aider.
  • For agents on cloud machines you check from a phone, use Amp ($0 to $200).
  • For a hosted app rather than a repository, use Replit Agent Core ($20).

OpenClaw, a free assistant you host yourself and talk to from chat apps, is a different kind of agent: OpenClaw vs Claude Code compares the two.

What developers report

  • "dozens of open source Claude code alternatives that are as good or better" (a Hacker News commenter, 17 Oct 2026, link)
  • "out-of-the-box Claude Code or Codex are likely getting the most cutting edge features" (a Hacker News commenter, 4 Nov 2026, link)
  • "Google Antigravity CLI using Gemini Flash 3.8 most of the time." (a Hacker News commenter, 4 Nov 2026, link)
  • A commenter recommending OpenCode noted that its Go plan works with any harness, so the Claude CLI can stay while the models change (Hacker News, 1 Nov 2026, link).
  • A $10 OpenCode Go subscriber running mostly DeepSeek Flash said the allowance had never run out (Hacker News, 4 Nov 2026, link).
  • A subscriber on the $100 Copilot tier was looking for a lower-cost plan and found OpenRouter more expensive for flagship models (Hacker News, 4 Nov 2026, link).

Using Aight with an alternative agent

Claude Code speaks Anthropic's Messages API and Codex CLI the Responses API, so neither works with Aight, which documents an OpenAI-compatible Chat Completions API. Agents with an OpenAI-compatible provider can be configured to use it: OpenCode, Cline, Kilo Code, Copilot CLI, Amp's custom URL and Aider. Aight's API reference documents model, messages and stream, and most agents also send tool definitions, so test tool calls before relying on them. Aider fits best because it reads edits as text.

export OPENAI_API_BASE=https://api.getaight.ai/v1
export OPENAI_API_KEY=sk-aight-...
aider --model openai/claude-sonnet-5-5

Configured this way, one key reaches Claude Sonnet 5.5 and GPT-6.1 Sol at list price ($2 / $10), and $103 buys $110 of balance outside India, so the daily month above costs $132.14 instead of $141.12. That still loses to a Claude Max plan at $100 for daily Claude work, and DeepSeek V4.1 Flash costs $7.47 direct off-peak against $14.00 on Aight. What Aight adds is a monthly cap and an allowed-model list per key. Each model's context window and maximum output are on the models page.

FAQ

Is there a free alternative to Claude Code?

Yes. Antigravity CLI has a free weekly quota on Gemini 3.8 Flash, and OpenCode includes free models. Cline, Kilo Code and Aider cost nothing themselves; you pay only the model provider, or nothing with a local model through Ollama.

Can Claude Code use GPT, Gemini or DeepSeek models?

Not with Anthropic's support: its docs say it doesn't support routing Claude Code to non-Claude models through any gateway. For other models, use OpenCode, Cline or Kilo Code.

What is the best open-source Claude Code alternative?

In the terminal, OpenCode (MIT), which runs any provider and has a desktop app. In VS Code, Cline (Apache-2.0) or Kilo Code (MIT), which also covers JetBrains. Codex CLI is open source too, but custom models must speak OpenAI's Responses API.

Is Gemini CLI still free?

Not for individuals. Since 18 June 2026, free and Google AI Pro or Ultra users get Antigravity CLI instead, and Gemini CLI serves Code Assist Standard and Enterprise licenses and paid API keys.

Can a Claude Pro or Max plan be used in another agent?

Some agents support it. Cline's Claude Code provider runs the official claude CLI on your plan, and Amp's Claude Code mode uses Anthropic's Agent SDK. Anthropic's Agent SDK docs say third-party products can't offer claude.ai login unless previously approved, so check each tool's terms.

Does Codex cost less than Claude Code?

Both start at $20 a month. Per task at API prices, Codex with GPT-6.1 Sol cost $1.50 against $11.42 for Claude Code with Opus 5.5 on Artificial Analysis's index, at a lower score.


Sources

Checked on 5 November 2026. Updated 17 November 2026: Aight's current model list; Claude Code 2.1.326 changes nothing else compared here.