AI Coding Agents Ranking (Oct 2026)
Last checked: 2026-10-21. We don't compute a score of our own. The order comes from public leaderboards, cited below with their dates. Product facts link to official pages read on 2026-10-21. "Unverified" means we couldn't confirm a value. We haven't tested these tools hands-on.
Machine-readable version: ai-coding-agents.json.
Main table
How it's ordered: first by Terminal-Bench 4.0, the only one of these boards that tests the products themselves (agent + model). Tools with no entry come next, ordered by the LMArena WebDev Elo of the best model the product officially offers. That's a model score, not a product score. Tools with no citable model are listed last as unranked, alphabetically. Equal scores share a rank (=4).
| Order | Tool | Terminal-Bench 4.0 (product) | WebDev Arena Elo (model, not product) | Entry paid plan | Free tier |
|---|---|---|---|---|---|
| 1 | Claude Code | 64.85% | 1807 (Claude Opus 5.5 (Max)) | $20/mo | No |
| 2 | OpenAI Codex | 58.18% | 1774 (GPT-6 Astra (Max)) | $20/mo | Yes |
| 3 | Kiro | no entry | 1689 (Claude Opus 5 (Max)) | $20/mo | Yes |
| =4 | Gemini CLI / Antigravity CLI | no entry | 1440 (Gemini 3 Pro) | Unverified | Unverified |
| =4 | Jules | no entry | 1440 (Gemini 3 Pro) | Unverified | Yes |
| unranked | Aider | no entry | depends on model | $0 (BYOK) | Yes |
| unranked | Cline | no entry | depends on model | $0 (BYOK) | Yes |
| unranked | Cursor | no entry | Unverified | $8/mo | Yes |
| unranked | Devin (incl. former Windsurf, now Devin Desktop) | no entry | depends on model | $20/mo | Yes |
| unranked | GitHub Copilot | no entry | Unverified | $10/mo | Yes |
| unranked | Kilo Code | no entry | depends on model | $0 (BYOK) | Yes |
Cited leaderboards
| Leaderboard | Measures | Data as of | Used for ordering? |
|---|---|---|---|
| Terminal-Bench 4.0 | Products (agent + model) | board updated 2026-10-06T23:20Z (2026-10-07 07:20 Beijing time) | Yes, first key |
| LMArena WebDev | Models | 2026-10-19 (903,893 votes, 149 models) | Yes, second key |
| SWE-bench | Models + scaffolds | newest entry dated 2026-02-26: STALE, no 2026 frontier models | No: data is stale |
| Aider Polyglot | Models | newest entry dated 2025-10-03: STALE | No: data is stale |
Terminal-Bench 4.0, top product runs: Claude Code + Opus 5.5 (max) 64.85% ±3.1, 2026-09-22 · Claude Code + Sonnet 5.5 (max) 61.8% ±2.9, 2026-09-28 · Codex + GPT-6 Astra (max) 58.18% ±2.8, 2026-09-03 · Codex + GPT-6.1 Sol (max) 58.18% ±3.1, 2026-09-29. None of the other products has an entry.
WebDev Arena (2026-10-19), models used above: Claude Opus 5.5 (Max) 1807 · GPT-6 Astra (Max) 1774 · Claude Opus 5 (Max) 1689 · Gemini 3 Pro 1440.
Verified product facts
| Tool | Paid tiers | Local | Cloud | Mobile | Repo required? | Models | Status |
|---|---|---|---|---|---|---|---|
| Claude Code | Free $0 (no Claude Code) · Pro $20 ($17 annual) incl. Claude Code · Max from $100 · Team/Enterprise | Yes | Yes | Yes: Claude app Code tab (cloud), Remote Control, Dispatch | Cloud sessions: yes (GitHub repo). Local: no | Anthropic Claude models | Active |
| OpenAI Codex | Included in ChatGPT Free/Go/Plus/Pro/Business/Edu/Enterprise; Plus $20 (Go price Unverified) | Yes | Yes | Yes: ChatGPT mobile / iOS | Cloud: yes (GitHub). CLI: no | OpenAI models | Active |
| Kiro | Free · Pro $20 · Pro+ $40 · Pro Max $100 · Power $200; add-on $0.04/credit | Yes | Yes | Unverified | Unverified | Auto, GPT-5.6, Claude Opus 5/Sonnet 5 etc., DeepSeek, MiniMax, GLM-5, Qwen3 Coder | Active |
| Gemini CLI / Antigravity CLI | Unverified | Yes | No | No (terminal tool) | No | Gemini 3 (Gemini 4 Argon needs Antigravity) | Gemini CLI replaced by Antigravity CLI for unpaid & Google One users on 2026-06-18 |
| Jules | Free 15 tasks/day · tier 100 tasks/day · tier 300 tasks/day (prices not shown) | No | Yes | Unverified | Yes (GitHub repo + branch) | Gemini 2.5 Pro / Gemini 3 Pro | Active |
| Aider | Free (BYOK) | Yes | No | No | No (git integration optional) | Any (OpenAI, Anthropic, Gemini, DeepSeek, Ollama, OpenRouter…) | Active |
| Cline | Open Source free · Enterprise (custom) | Yes | No | No | No | Any (BYOK: Anthropic, Gemini, OpenAI, OpenRouter, Bedrock, Vertex, DeepSeek…) | Active |
| Cursor | Free $0 · Plus $8 (Grok models only) · Premium $30 · more | Yes | Yes | Yes: Cursor iOS app, cursor.com/agents PWA | No ('Start from scratch' creates a draft repo; paid plan) | Multiple vendors incl. third-party models | Active |
| Devin (incl. former Windsurf, now Devin Desktop) | Free $0 · Pro $20 · Max $200 · Teams $80 + $40/full seat · Enterprise | Yes | Yes | Unverified | Unverified | OpenAI, Claude, Gemini, SpaceXAI, open-source, SWE-2 | Active; Windsurf renamed Devin Desktop on 2026-06-02; windsurf.com/pricing redirects to devin.ai/pricing |
| GitHub Copilot | Free $0 · Pro $10 · Pro+ $39 · Max $100 | Yes | Yes | Unverified | Cloud agent: yes (GitHub repo) | Multi-model | Active |
| Kilo Code | Free (individuals) · Teams $15/user · Enterprise; inference via Auto Free/BYOK/local, Kilo Gateway at provider rates, or Kilo Pass | Yes | Yes | Unverified | Unverified | Any (500+ models across 60+ providers, BYOK, local) | Active |
Sorted: cheapest to start
| Tool | Entry paid plan | Free tier |
|---|---|---|
| Aider | 0 | Yes |
| Cline | 0 | Yes |
| Kilo Code | 0 | Yes |
| Cursor | 8 | Yes |
| GitHub Copilot | 10 | Yes |
| Claude Code | 20 | No |
| OpenAI Codex | 20 | Yes |
| Kiro | 20 | Yes |
| Devin (incl. former Windsurf, now Devin Desktop) | 20 | Yes |
| Gemini CLI / Antigravity CLI | Unverified | Unverified |
| Jules | Unverified | Yes |
AI code review (separate, not ranked)
| Tool | Plans | Free tier | Where it reviews |
|---|---|---|---|
| CodeRabbit | Essentials $24 · Team $48 · Advanced $72 /user/mo · Enterprise (billing period Unverified: page has a monthly/annual toggle) | Unverified (14-day trial shown) | PRs, CLI, IDE, agent loops |
None of the cited leaderboards measures code review, so this section isn't ranked.
Affiliate disclosure: we have no affiliate links on this page. The order comes only from the cited data.
For how to drive these from a phone, see our mobile AI coding agents guide.
Unverified items
- Claude Code · models: Third-party models listed on Terminal-Bench (e.g. GLM-5.3) imply API redirection, not verified on official docs
- Kiro · mobile: Kiro Web exists; mobile use not confirmed
- Kiro · repo_required
- Gemini CLI / Antigravity CLI · free_tier: Free tier now via Antigravity CLI; limits not verified (third-party says 1,000 req/day, not official)
- Gemini CLI / Antigravity CLI · entry_paid_usd
- Gemini CLI / Antigravity CLI · tiers
- Jules · entry_paid_usd: Paid tiers (100 and 300 tasks/day) shown without prices
- Jules · mobile
- Cursor · model_score_webdev
- Devin (incl. former Windsurf, now Devin Desktop) · mobile: Slack/Teams integrations listed; no native mobile app confirmed
- Devin (incl. former Windsurf, now Devin Desktop) · repo_required: GitHub/GitLab/Bitbucket integrations listed
- GitHub Copilot · mobile: Docs nav lists 'Chat in Mobile'; assigning cloud-agent tasks from mobile not confirmed today
- GitHub Copilot · model_score_webdev
- Kilo Code · mobile
- Kilo Code · repo_required
- CodeRabbit: whether there is a free plan, and whether the listed prices are monthly or annual.
- Gemini CLI model mapping: the official page says "Gemini 3" without naming the variant, so we assumed Gemini 3 Pro.
- Amazon Q Developer: we didn't check whether it is still sold; Kiro is AWS's agent product in this list.
Get the 5 best new AI tools every week
One short email every Friday. No spam, unsubscribe anytime.