Loading the page.
AI agents
all AI agents for coding in one table: live benchmark scores, prices, and formats. Data with dates and sources, updated every Tuesday. Methodology
index_ · measurement #10 from 09/22/2026
Leader — Codex · 253
Next measurement 09/29/2026 · methodology
Vibecoding news
- OpenCode Desktop Is Already Used by Millions: 2M Weekly Active Users09/24
- Review GitHub PRs through Muse without switching tabs09/23
- A coding setup speeds up work without extra editor time: Harry Tan's experience09/23
Analysis · 09/21.2026
The Jev model doesn't write text, it judges: we put it in charge of our news feed
All articles
47 agents in the table · segment ~410
| # | Agent | index_ ↓ | Trend | RuBench | Entry | Models | From Russia | Code | Hosted in |
|---|---|---|---|---|---|---|---|---|---|
| 1 |  Codex· OpenAI | 253 | −7,1 | 82,6GPT-5.6 Sol · xhigh | $0 in ChatGPT | own GPT | IP + card | Open | terminaleditorclouddesktop |
| 2 |  Claude Code· Anthropic | 208 | −3,9 | 85,0Fable 5 · xhigh | from $17 | own Claude | IP + card | Closed | terminaleditorclouddesktop |
| 3 |  Cursor· Anysphere | 187 | −9,0 | — | $0 · Pro $20 | any + Composer | foreign card | Closed | terminalcloudeditor |
| 4 |  GitHub Copilot· GitHub | 30 | +0,1 | — | $0 · Pro $10 | your choice + key | foreign IP | Closed | cloudterminaleditor |
| 5 |  Antigravity· Google | 27 | +0,1 | 47,2Gemini 3.1 Pro · high | $0 preview | your choice | foreign IP | Closed | browsereditor |
| 6 |  Replit· Replit | 26 | −5,3 | — | $0 · Core $18 | external | foreign card | Closed | cloudbuilder |
| 7 |  Lovable· Lovable | 23 | −1,2 | — | $0 · Pro $25 | external | foreign card | Closed | builder |
| 8 |  OpenCode· SST | 18 | −1,9 | — | $0, you pay for the model | BYOK | no restrictions | Open | terminal |
| 9 |  Cline· Cline | 14 | −0,1 | — | $0, you pay for the model | BYOK | no restrictions | Open | editor |
| 10 |  DeepSeek Harness· DeepSeek | 13 | −0,8 | — | — | — | — | Open | terminal |
| 11 |  Windsurf· Cognition | 8 | −0,8 | — | — | — | — | Closed | editor |
| 12 |  Kiro· AWS | 8 | −0,4 | — | — | — | — | Closed | editor |
| 13 |  Framer· Framer | 7 | −0,2 | — | — | — | — | Closed | builder |
| 14 |  Qwen Code· Alibaba | 7 | −0,4 | — | — | — | — | Open | terminal |
| 15 |  Grok Build· SpaceXAI | 7 | −0,1 | — | $0 in Grok chat | own grok-4.6 | IP + card | Open | terminal |
| 16 |  Devin· Cognition | 6 | +0,3 | — | — | — | — | Closed | cloud |
| 17 |  Manus· Butterfly Effect | 5 | −0,4 | — | $0 · from $20 | external | IP + card | Closed | cloudbrowser |
| 18 |  Genspark· MainFunc | 4 | +0,1 | — | — | — | — | Closed | browser |
| 19 |  Gemini CLI· Google | 3 | −0,8 | — | $0 with Google login | own Gemini | foreign IP | Open | terminal |
| 20 |  Trae· ByteDance | 3 | +0,2 | — | — | — | — | Closed | editor |
| 21 |  Augment Code· Augment Code | 3 | −0,4 | — | — | — | — | Closed | terminalcloud |
| 22 |  Aider· open source | 2 | −1,9 | — | — | — | — | Open | terminal |
| 23 |  Bolt· StackBlitz | 2 | +0,5 | — | $0 · Pro $20 | Claude | IP + card | Closed | builder |
| 24 |  Comet· Perplexity | 1 | −0,1 | — | — | — | — | Closed | browser |
| 25 |  Roo Code· Roo Code | 1 | −0,4 | — | — | — | — | Open | editor |
| 26 |  Base44· Wix | 1 | 0,0 | — | — | — | — | Closed | builder |
| 27 |  Qoder· Alibaba | 1 | 0,0 | — | — | — | — | Closed | desktopeditor |
| 28 |  GLM Coding· Z.ai | 1 | 0,0 | — | — | — | — | Closed | cloud |
| 29 |  Zed· Zed Industries | 1 | −0,2 | — | — | — | — | Open | editor |
| 30 |  Junie· JetBrains | 1 | +0,1 | — | — | — | — | Closed | editor |
| 31 |  ChatGPT Agent· OpenAI | 1 | 0,0 | — | — | — | — | Closed | cloudbrowser |
| 32 |  Emergent· Emergent | 1 | −0,3 | — | — | — | — | — | builder |
| 33 |  OpenHands· All Hands AI | 1 | +0,1 | — | — | — | — | Open | terminalbrowsercloud |
| 34 |  Runner H· H Company | 1 | 0,0 | — | — | — | — | Closed | browsercloud |
| 35 |  Warp· Warp | 1 | 0,0 | — | — | — | — | Open | desktopcloudterminal |
| 36 |  Dia· The Browser Company | 1 | −0,1 | — | — | — | — | Closed | browser |
| 37 |  Rork· Rork | 0 | +0,1 | — | — | — | — | — | builder |
| 38 |  v0· Vercel | 0 | −0,5 | — | — | — | — | Closed | builder |
| 39 |  Firebase Studio· Google | 0 | 0,0 | — | — | — | — | Closed | cloudbuilder |
| 40 |  Amp· Sourcegraph | 0 | 0,0 | — | — | — | — | Closed | editorterminal |
| 41 |  Goose· Linux Foundation | 0 | −0,5 | — | — | — | — | Open | desktopterminal |
| 42 |  Opera Neon· Opera | 0 | 0,0 | — | — | — | — | Closed | browser |
| 43 |  Jules· Google | 0 | 0,0 | — | — | — | — | Closed | cloud |
| 44 |  Pi· Earendil | 0 | −0,2 | — | — | — | — | Open | terminal |
| 45 |  Droid· Factory | 0 | 0,0 | — | — | — | — | Closed | editorcloudterminal |
| 46 |  Crush· Charm | 0 | 0,0 | — | — | — | — | Open | terminal |
| 47 |  Databutton· Databutton | 0 | 0,0 | — | — | — | — | — | builder |
index_ · measurement #10 from 09/22/2026
Leader — Codex · 253
Next measurement 09/29/2026 · methodology
Vibecoding news
- OpenCode Desktop Is Already Used by Millions: 2M Weekly Active Users09/24
- Review GitHub PRs through Muse without switching tabs09/23
- A coding setup speeds up work without extra editor time: Harry Tan's experience09/23
Analysis · 09/21.2026
The Jev model doesn't write text, it judges: we put it in charge of our news feed
All articles
More on the topic
Tests
- Code agent ranking
- RuBench: agents on Russian technical specs
- How to read code benchmarks
Money
- Subscription multiplier: where $20 gets $600
- Pay for AI from Russia
How to read the table
An agent - when you hand off a task and leave, but the work continues: it plans steps, edits files, checks the result, and delivers it done. A chat assistant answers a question - an agent sees the task through.
Formats: where the agent lives
- desktop - Its own window on your computer: sessions and tasks, no terminal or code
- terminal - Lives in the console, digs through the whole project
- editor - An agent inside the IDE, code in plain sight
- builder - Describe it - get a site; the code is hidden
- cloud - Background worker: give it a task and walk away
- browser - Browses the web, does the whole task
Dark tag in the table - native format: the agent is designed for it.
Numbers: what to trust
- vibecoding.tech Index - agent visibility in the Russian-speaking market: search queries, mentions, job listings; the tables are sorted by it. Methodology
- RuBench - our benchmark on Russian technical specs: pass@1 with open trajectories, under the score - agent+model pair and run date. Empty - the agent has not run yet; new models enter in the next round.
Table notes
- Index - rounded scores from measurement No. 10 on 09/22/2026; score change vs. measurement #9 from 09/17/2026.
- new - newcomers to the registry as of 09/17/2026: baseline using the same formula, change from the next measurement.
- RU agents - in the overall ranking, index on the same scale.
- «Code -» = openness not verified.
- Login, Models, and From Russia - details from the agent card, by clicking the row.
- BYOK (bring your own key) - the agent is free and works with your API key from the selected provider.
- Fable 5's row in RuBench - with safeguard fallback: 5 of 25 tasks made it to Opus 4.8; analysis - on the RuBench page.
- Agentic code reviewers - in their own table Code review.
Source of scores - RuBench, round 2 (r2): methodology and trajectories
Friday email
Once a week: what changed in AI and what it means for you. Three minutes, numbers with sources.