Claude Code (9.0) is the most capable tool if you’re willing to pay for a subscription. Kilo Code is the free, open-source pick with zero lock-in — you bring your own model or use free ones at no cost. Codex CLI is the best choice if you already pay for ChatGPT, since it’s included across all plans starting at $0.
How We Tested
This review is based on official documentation, pricing pages, and published third-party reports — we did not run the tool hands-on. We pulled facts from each tool’s official product and pricing pages, GitHub license files, and the Vals AI SWE-bench Verified leaderboard. We also checked published third-party reports on the March 2026 Claude Code source leak. We did not verify actual response quality, real-world speed, or hands-on usability — so treat this as a research-backed comparison, not a hands-on test.
Kilo Code vs Claude Code vs Codex CLI: the short version
- Kilo Code — Best for: beginners who want a free, open-source coding agent with zero lock-in and 500+ model options.
- Claude Code — Best for: developers who want the single most capable coding agent and already pay for Claude (claude.com/product/claude-code).
- Codex CLI — Best for: ChatGPT subscribers who want a powerful terminal agent without an extra subscription (openai.com/chatgpt/pricing/).
| What it is | Best for | Entry price | Free tier | Open source | Best-model SWE-bench | Context window |
|---|---|---|---|---|---|---|
| Free, open-source AI coding assistant (VS Code, JetBrains, CLI) | No lock-in, 500+ models, free or BYOK | $0 | Yes (free models / BYOK) | ✅ MIT | No official harness score | Model-dependent |
| Anthropic’s closed-source coding agent (terminal, editor, desktop, web) | Most capable agent, deep Claude integration | $20/mo (Pro) | No (Pro required) | ❌ Proprietary | Claude Opus 5: 97.00% | 200k (chat plans) |
| OpenAI’s terminal coding agent (CLI, IDE, ChatGPT) | ChatGPT subscribers, cloud features | $0 (limited) | Yes (limited Codex) | ✅ Apache-2.0 | GPT-5.6 Sol: 96.20% | Up to 400k (Pro) |
Pricing and licensing from each vendor’s official page: Kilo Code pricing, Claude Code, ChatGPT pricing. Benchmark scores from the Vals AI SWE-bench Verified leaderboard.
Scores at a glance
| Dimension | Kilo Code | Claude Code | Codex CLI |
|---|---|---|---|
| Overall Score | 7.9/10 | 9.0/10 | 8.5/10 |
| Ease of Use | 8.5/10 | 8.5/10 | 8.0/10 |
| Features | 8.0/10 | 9.5/10 | 9.0/10 |
| Performance | 7.5/10 | 9.5/10 | 9.0/10 |
| Documentation | 8.0/10 | 9.0/10 | 8.5/10 |
| Support | 7.5/10 | 8.5/10 | 8.0/10 |
Scores from the ToolBrain comparison database → /comparisons/.
What each tool is
Kilo Code
Kilo Code is a free, open-source AI coding assistant that runs inside your code editor (VS Code or JetBrains) or your terminal. Unlike tools that lock you into one AI model, Kilo Code lets you pick from 500+ models, switch mid-task, and pay the model provider’s price with zero markup — or bring your own API keys. It’s the same agent everywhere: editor, CLI, and cloud agents. Kilo Code is model-agnostic, meaning capability tracks whichever model you run. It supports BYOK (bring your own key), so you can use your existing Anthropic, OpenAI, Google, or AWS Bedrock accounts. It also supports MCP (Model Context Protocol), which lets you connect external tools and data sources to the agent. Kilo Code uses worktrees — isolated environments where the agent can make changes without affecting your main codebase.
Claude Code
Claude Code is Anthropic’s AI coding agent that works from your terminal, editor, or desktop app. You describe a task in plain English and it reads your codebase, edits files, runs commands, and handles git workflows on its own. It’s the tool Anthropic built its own engineering team on and is widely considered the strongest agentic coding tool on the market. Claude Code supports MCP, worktrees, and subagents — lightweight agents that handle specific parts of a larger task. It’s closed source, with its license stating ”© Anthropic PBC. All rights reserved” github.com/anthropics/claude-code. On March 31, 2026, the full Claude Code source accidentally leaked via an npm packaging error — a 59.8 MB source-map file with ~512,000 lines of unobfuscated TypeScript. Anthropic called it “an honest mistake,” not a breach, but the code was publicly mirrored and forked within hours onehack.st.
Codex CLI
Codex CLI is OpenAI’s coding agent that runs in your terminal — you type a task and it reads your project, edits files, runs commands, and can open pull requests. The same Codex agent also lives inside ChatGPT (web/desktop) and as an IDE extension, all connected to your ChatGPT account. The CLI itself is open source under Apache-2.0; the GPT-5.6 models behind it are closed. Codex CLI supports MCP, worktrees, subagents, and sandboxed execution. It can connect to GitHub for automated code review and pull request handling. You can run it with your own API key for pay-per-token usage, or sign in with your ChatGPT account to access cloud features.
Pricing compared for beginners
Kilo Code starts at $0 for individuals — you can use free models, bring your own API keys, or run local models at no cost kilo.ai/pricing. Teams pay $15/user/month. If you want premium models through Kilo’s Gateway, a Kilo Pass starts at $19/month. Cloud compute (like hosted dev environments) is billed separately at $0.33–$1.20/hour.
Claude Code is included with Claude Pro ($20/month) and Max plans ($100–$200/month) claude.com/product/claude-code. The free plan does not include Claude Code. If you use the API instead, Opus 5 costs $5–$10 input per million tokens and $25–$50 output per million tokens.
Codex CLI is included across all ChatGPT plans, starting with Free at $0 (limited access) openai.com/chatgpt/pricing/. Go is $8/month, Plus is $20/month, and Pro starts at $100/month for higher rate limits. Business is $25/user/month. You can also use an API key to pay only for tokens used.
Features compared
Models: Kilo Code supports 500+ models across 60+ providers, with auto-routing between efficient, frontier, and free tiers. Claude Code uses Claude Opus 5, Sonnet, Haiku, and Fable claude.com/product/claude-code. Codex CLI uses GPT-5.6 Sol (frontier), Terra (mid-tier), and Luna (fast/cheap), plus GPT-5.6 Sol Pro on Pro plans openai.com/chatgpt/pricing/.
Context windows: Kilo Code’s context depends on which model you choose. Claude Code offers 200k tokens on chat plans claude.com/product/claude-code. Codex CLI supports up to 400k tokens on Pro, 256k on Go/Plus, and 128k on the GPT Instant tier openai.com/chatgpt/pricing/.
Platforms: Kilo Code works in VS Code, JetBrains, CLI, Slack, and cloud agents. Claude Code runs in terminal, VS Code, JetBrains, the Claude desktop app, web, Slack, GitHub Actions, and mobile. Codex CLI runs in terminal, IDE extension, ChatGPT web/desktop, and cloud environments.
Agentic features: All three support MCP, worktrees, and subagents. Claude Code adds dynamic workflows (parallel subagents), agent view, routines (scheduled automation), and computer use. Codex CLI adds sandboxed execution, image input, web search, and scheduled tasks. Kilo Code adds browser automation and custom modes (Ask, Architect, Code, Debug).
Benchmarks compared
On the Vals AI SWE-bench Verified leaderboard (an independent benchmark using 500 human-validated GitHub issues), Claude Opus 5 scores 97.00% — the highest of any model tested vals.ai/benchmarks/swebench. GPT-5.6 Sol scores 96.20%, and GPT-5.6 Luna scores 93.00%. Multiple aggregators report consistent numbers for these models.
Kilo Code has no official SWE-bench harness score. Because it’s a model-agnostic harness — it works with any model you plug in — its performance depends entirely on which model you choose. Kilo publishes a model leaderboard at kilo.ai/leaderboard, but these are model-level scores, not harness-level scores.
Open-source status
Kilo Code is open source under the MIT license, with 26.9k stars on GitHub github.com/Kilo-Org/kilocode. Codex CLI is open source under Apache-2.0, with 106.5k stars github.com/openai/codex. Claude Code is not open source — its GitHub repo contains only docs, scripts, plugins, and issue tracking, with a LICENSE.md that reads ”© Anthropic PBC. All rights reserved” github.com/anthropics/claude-code. Anthropic’s 2025 plan to open-source Claude Code has not materialized as of August 2026.
Beginner takeaway
- If you want to try a coding agent for free with no risk: start with Kilo Code or Codex CLI on the Free plan.
- If you already pay for ChatGPT Plus or Pro: use Codex CLI — it’s included.
- If you want the most capable tool and don’t mind paying $20/month: choose Claude Code (Pro).
- If you care about open source and zero lock-in: Kilo Code is the only fully open-source option with a free tier.
Who should use this
Use Kilo Code if you want a free, open-source coding agent and don’t mind configuring your own models or API keys. It’s ideal if you switch between AI providers or want to avoid vendor lock-in. Don’t use Kilo Code if you want a polished, out-of-the-box experience with a single best-in-class model — you’ll need to choose and pay for models separately.
Use Claude Code if you want the most capable coding agent available and already use or are willing to subscribe to Claude. It offers the best performance and deepest integration. Don’t use Claude Code if you require open-source software or are concerned about the March 2026 source leak exposure.
Use Codex CLI if you already pay for ChatGPT and want a powerful terminal agent without an extra subscription. The Free and Go tiers let you try it at no additional cost. Don’t use Codex CLI if you don’t use ChatGPT and want to avoid rate limits — you’ll need to upgrade to Plus or Pro for serious usage.
Related comparisons
- AI Coding Agent comparison
- ToolBrain comparison database
- ToolBrain roadmap
- Cron pipeline
- our Kilo Code review
- our earlier Claude Code vs Codex CLI comparison
- our Claude Code review
📊 See how these tools compare → /comparisons/
Back to all posts
