clawdboard_
← all stats

>AI Coding Tool Comparison

Side-by-side comparison of Claude Code, Codex CLI, OpenCode, Claude Code Desktop, OpenCode Go, Hermes Agent, DeepSeek Harness, GitHub Copilot CLI, Cursor, Pi, Antigravity, and Gemini CLI usage from 84 developers on clawdboard. All costs are estimated from token counts and published API pricing.

As of October 11, 2026 at 2:06 AM UTC, 84 developers have tracked $1.27M in estimated AI coding spend and 2019.3B tokens across 12 tools on clawdboard. Claude Code leads with 73.5% of total spend ($933k), followed by Codex CLI at 21.7% ($276k), followed by OpenCode at 4.4% ($56.3k), followed by Claude Code Desktop at 0.2% ($3.1k), followed by OpenCode Go at 0.0% ($511.53), followed by Hermes Agent at 0.0% ($429.14), followed by DeepSeek Harness at 0.0% ($426.24), followed by GitHub Copilot CLI at 0.0% ($346.88), followed by Cursor at 0.0% ($135.54), followed by Pi at 0.0% ($92.50), followed by Antigravity at 0.0% ($41.20), and Gemini CLI at 0.0% ($8.93). Data is updated hourly from opt-in developer usage logs.

Last updated: October 11, 2026 at 2:06 AM UTC · Refreshed hourly

>Community Totals

Aggregate usage across all tracked AI coding tools.

Total Estimated Cost

$1.27M

across all tools

Total Tokens

2019.3B

processed across all tools

Tools Tracked

12

AI coding tools

>Cost Share by Tool

How is AI coding spend distributed across tools?

Claude Code — 73.5% ($933k)
Codex CLI — 21.7% ($276k)
OpenCode — 4.4% ($56.3k)
Claude Code Desktop — 0.2% ($3.1k)
OpenCode Go — 0.0% ($511.53)
Hermes Agent — 0.0% ($429.14)
DeepSeek Harness — 0.0% ($426.24)
GitHub Copilot CLI — 0.0% ($346.88)
Cursor — 0.0% ($135.54)
Pi — 0.0% ($92.50)
Antigravity — 0.0% ($41.20)
Gemini CLI — 0.0% ($8.93)

>Tool Breakdown

Usage, cost, and adoption metrics for each tracked tool.

Claude Code

Anthropic's official CLI for Claude. An agentic coding assistant that works directly in your terminal with full access to your codebase.

73.5%

Estimated Cost

$933k

73.5% of total

Total Tokens

1426.1B

14.4B in / 3.6B out

Developers

52

avg: $17.9k / med: $5.1k

Active Days

5,083

since April 28, 2025

Claude Code accounts for 73.5% of community spend on clawdboard with $933k in estimated API cost across 52 developers. The average Claude Code user has spent an estimated $17.9k (median: $5.1k), consuming 1426.1B tokens (14.4B input, 3.6B output). The most-used models through Claude Code are Opus 5 (26.5% of Claude Code spend), Opus 5.5 (17.4% of Claude Code spend), and Opus 4.8 (16.4% of Claude Code spend). Usage has been tracked since April 28, 2025.

Codex CLI

OpenAI's command-line coding agent that uses GPT-4o and o-series models. Brings OpenAI's models to the terminal for code generation and editing.

21.7%

Estimated Cost

$276k

21.7% of total

Total Tokens

484.9B

12.2B in / 1.2B out

Developers

33

avg: $8.3k / med: $378.93

Active Days

1,367

since September 17, 2025

Codex CLI accounts for 21.7% of community spend on clawdboard with $276k in estimated API cost across 33 developers. The average Codex CLI user has spent an estimated $8.3k (median: $378.93), consuming 484.9B tokens (12.2B input, 1.2B output). The most-used models through Codex CLI are gpt-5.6-sol (45.9% of Codex CLI spend), gpt-daybreak-blue-latest (19.7% of Codex CLI spend), and gpt-6-astra (16.3% of Codex CLI spend). Usage has been tracked since September 17, 2025.

OpenCode

An open-source terminal-based AI coding assistant that supports multiple LLM providers. Designed as a flexible alternative with provider-agnostic model support.

4.4%

Estimated Cost

$56.3k

4.4% of total

Total Tokens

89.5B

9.9B in / 562.3M out

Developers

13

avg: $4.3k / med: $24.82

Active Days

365

since August 8, 2025

OpenCode accounts for 4.4% of community spend on clawdboard with $56.3k in estimated API cost across 13 developers. The average OpenCode user has spent an estimated $4.3k (median: $24.82), consuming 89.5B tokens (9.9B input, 562.3M output). The most-used models through OpenCode are gpt-5.6-sol (12% of OpenCode spend), Opus 5.5 (11.9% of OpenCode spend), and grok-4.6 (11.8% of OpenCode spend). Usage has been tracked since August 8, 2025.

Claude Code Desktop

The Claude desktop app's in-app coding agent (Cowork / Dispatch). Captures usage from local-agent-mode sessions running inside the desktop app, separate from the Claude Code CLI.

0.2%

Estimated Cost

$3.1k

0.2% of total

Total Tokens

2.6B

5.4M in / 467.8k out

Developers

17

avg: $185.21 / med: $24.87

Active Days

286

since January 1, 1999

Claude Code Desktop accounts for 0.2% of community spend on clawdboard with $3.1k in estimated API cost across 17 developers. The average Claude Code Desktop user has spent an estimated $185.21 (median: $24.87), consuming 2.6B tokens (5.4M input, 467.8k output). The most-used models through Claude Code Desktop are Fable 5 (27.3% of Claude Code Desktop spend), Opus 4.7 (26.5% of Claude Code Desktop spend), and Opus 4.6 (23.6% of Claude Code Desktop spend). Usage has been tracked since January 1, 1999.

OpenCode Go

OpenCode's Go subscription tier — a curated set of high-performance open-source models (GLM, MiMo, DeepSeek, Kimi, Qwen, MiniMax) accessed through the OpenCode TUI.

0%

Estimated Cost

$511.53

0% of total

Total Tokens

13.0B

285.6M in / 14.5M out

Developers

4

avg: $127.88 / med: $28.62

Active Days

39

since April 28, 2026

OpenCode Go accounts for 0% of community spend on clawdboard with $511.53 in estimated API cost across 4 developers. The average OpenCode Go user has spent an estimated $127.88 (median: $28.62), consuming 13.0B tokens (285.6M input, 14.5M output). The most-used models through OpenCode Go are glm-5.2 (81% of OpenCode Go spend), deepseek-v4-pro (10.2% of OpenCode Go spend), and deepseek-v4-flash (2.8% of OpenCode Go spend). Usage has been tracked since April 28, 2026.

Hermes Agent

Nous Research's self-improving CLI agent with persistent memory and 300+ supported models. Interactive sessions (terminal, desktop app, editor integrations, web UI) are tracked; scheduled cron jobs and chat gateways like Telegram, Discord, and WhatsApp are excluded.

0%

Estimated Cost

$429.14

0% of total

Total Tokens

557.9M

196.5M in / 3.4M out

Developers

5

avg: $85.83 / med: $1.17

Active Days

37

since April 8, 2026

Hermes Agent accounts for 0% of community spend on clawdboard with $429.14 in estimated API cost across 5 developers. The average Hermes Agent user has spent an estimated $85.83 (median: $1.17), consuming 557.9M tokens (196.5M input, 3.4M output). The most-used models through Hermes Agent are glm-5.3-flash (24% of Hermes Agent spend), grok-4.6 (19.5% of Hermes Agent spend), and omni-deepseek-v4-flash (17.8% of Hermes Agent spend). Usage has been tracked since April 8, 2026.

DeepSeek Harness

DeepSeek's open-source agent runtime (dsh), where the model, tools, and session persistence are all swappable plugins. Defaults to DeepSeek models; other OpenAI-compatible endpoints also work.

0%

Estimated Cost

$426.24

0% of total

Total Tokens

1.4B

41.4M in / 4.5M out

Developers

2

avg: $213.12 / med: $213.12

Active Days

13

since August 15, 2026

DeepSeek Harness accounts for 0% of community spend on clawdboard with $426.24 in estimated API cost across 2 developers. The average DeepSeek Harness user has spent an estimated $213.12 (median: $213.12), consuming 1.4B tokens (41.4M input, 4.5M output). The most-used models through DeepSeek Harness are k3 (80.7% of DeepSeek Harness spend), deepseek-ai/DeepSeek-V4-Flash (16.5% of DeepSeek Harness spend), and deepseek-v4-flash-vision-exp (1.5% of DeepSeek Harness spend). Usage has been tracked since August 15, 2026.

GitHub Copilot CLI

GitHub's agentic command-line coding assistant. Multi-model support (Claude, GPT, etc.) billed in premium-request units alongside token-level metrics.

0%

Estimated Cost

$346.88

0% of total

Total Tokens

487.3M

34.4M in / 7.1M out

Developers

3

avg: $115.63 / med: $0.15

Active Days

25

since March 30, 2026

GitHub Copilot CLI accounts for 0% of community spend on clawdboard with $346.88 in estimated API cost across 3 developers. The average GitHub Copilot CLI user has spent an estimated $115.63 (median: $0.15), consuming 487.3M tokens (34.4M input, 7.1M output). The most-used models through GitHub Copilot CLI are claude-sonnet-4.6 (55.9% of GitHub Copilot CLI spend), gpt-5.4 (17.9% of GitHub Copilot CLI spend), and claude-haiku-4.5 (13.4% of GitHub Copilot CLI spend). Usage has been tracked since March 30, 2026.

Cursor

AI-first code editor built as a VS Code fork. Ships an in-editor chat panel and agent mode that work across your codebase, with usage tracked from the local SQLite store.

0%

Estimated Cost

$135.54

0% of total

Total Tokens

337.6M

321.1M in / 16.5M out

Developers

10

avg: $13.55 / med: $0.00

Active Days

284

since April 27, 2025

Cursor accounts for 0% of community spend on clawdboard with $135.54 in estimated API cost across 10 developers. The average Cursor user has spent an estimated $13.55 (median: $0.00), consuming 337.6M tokens (321.1M input, 16.5M output). The most-used models through Cursor are claude-4.5-sonnet-thinking (25.6% of Cursor spend), composer-1 (19.5% of Cursor spend), and claude-4.5-opus-high-thinking (15.8% of Cursor spend). Usage has been tracked since April 27, 2025.

Pi

A minimal, extensible terminal coding harness by Armin Ronacher and Mario Zechner. Provider-agnostic (Anthropic, OpenAI, Google, and more), with every session saved as a local JSONL log.

0%

Estimated Cost

$92.50

0% of total

Total Tokens

265.5M

6.8M in / 1.8M out

Developers

2

avg: $46.25 / med: $46.25

Active Days

20

since June 2, 2026

Pi accounts for 0% of community spend on clawdboard with $92.50 in estimated API cost across 2 developers. The average Pi user has spent an estimated $46.25 (median: $46.25), consuming 265.5M tokens (6.8M input, 1.8M output). The most-used models through Pi are deepseek-v4.1-flash (62.5% of Pi spend), muse-spark-1.3-contributor-free (31.3% of Pi spend), and deepseek-flash (1.5% of Pi spend). Usage has been tracked since June 2, 2026.

Antigravity

Google's agent-first IDE — a VS Code fork with deeply-integrated agentic workflows powered by Gemini 3 and other frontier models.

0%

Estimated Cost

$41.20

0% of total

Total Tokens

14.1M

12.7M in / 176.1k out

Developers

1

avg: $41.20 / med: $41.20

Active Days

52

since August 12, 2026

Antigravity accounts for 0% of community spend on clawdboard with $41.20 in estimated API cost across 1 developer. The average Antigravity user has spent an estimated $41.20 (median: $41.20), consuming 14.1M tokens (12.7M input, 176.1k output). The most-used models through Antigravity are gemini-3.6-flash-high (79.9% of Antigravity spend), gemini-3.5-flash-low (20% of Antigravity spend), and gemini-3.7-flash-high (0.1% of Antigravity spend). Usage has been tracked since August 12, 2026.

Gemini CLI

Google's official command-line coding agent powered by the Gemini family of models. Open source, with built-in multimodal and tool-use capabilities.

0%

Estimated Cost

$8.93

0% of total

Total Tokens

36.8M

6.7M in / 638.8k out

Developers

4

avg: $2.23 / med: $2.55

Active Days

9

since April 28, 2026

Gemini CLI accounts for 0% of community spend on clawdboard with $8.93 in estimated API cost across 4 developers. The average Gemini CLI user has spent an estimated $2.23 (median: $2.55), consuming 36.8M tokens (6.7M input, 638.8k output). The most-used models through Gemini CLI are gemini-3-flash-preview (63.4% of Gemini CLI spend), gemini-2.5-pro (27.5% of Gemini CLI spend), and gemini-3.1-pro-preview-customtools (5.9% of Gemini CLI spend). Usage has been tracked since April 28, 2026.

>Daily Cost Trends by Tool

How does daily estimated spend compare across Claude Code, Codex CLI, OpenCode, Claude Code Desktop, OpenCode Go, Hermes Agent, DeepSeek Harness, GitHub Copilot CLI, Cursor, Pi, Antigravity, and Gemini CLI? This chart shows the 7-day moving average to smooth out daily variance.

7-day moving average of estimated daily cost per tool

>Claude Code Model Breakdown

Which models do Claude Code users prefer? Cost share breakdown across all models used through Claude Code.

>Codex CLI Model Breakdown

Which models do Codex CLI users prefer? Cost share breakdown across all models used through Codex CLI.

>OpenCode Model Breakdown

Which models do OpenCode users prefer? Cost share breakdown across all models used through OpenCode.

>Claude Code Desktop Model Breakdown

Which models do Claude Code Desktop users prefer? Cost share breakdown across all models used through Claude Code Desktop.

>OpenCode Go Model Breakdown

Which models do OpenCode Go users prefer? Cost share breakdown across all models used through OpenCode Go.

>Hermes Agent Model Breakdown

Which models do Hermes Agent users prefer? Cost share breakdown across all models used through Hermes Agent.

>DeepSeek Harness Model Breakdown

Which models do DeepSeek Harness users prefer? Cost share breakdown across all models used through DeepSeek Harness.

>GitHub Copilot CLI Model Breakdown

Which models do GitHub Copilot CLI users prefer? Cost share breakdown across all models used through GitHub Copilot CLI.

>Cursor Model Breakdown

Which models do Cursor users prefer? Cost share breakdown across all models used through Cursor.

>Pi Model Breakdown

Which models do Pi users prefer? Cost share breakdown across all models used through Pi.

>Antigravity Model Breakdown

Which models do Antigravity users prefer? Cost share breakdown across all models used through Antigravity.

>Gemini CLI Model Breakdown

Which models do Gemini CLI users prefer? Cost share breakdown across all models used through Gemini CLI.

AI Coding Tool Landscape

The AI coding tool ecosystem is evolving rapidly. clawdboard currently tracks 12 tools: Claude Code (Anthropic), Codex CLI (OpenAI), OpenCode (Community), Claude Code Desktop (Anthropic), OpenCode Go (OpenCode), Hermes Agent (Nous Research), DeepSeek Harness (DeepSeek), GitHub Copilot CLI (GitHub), Cursor (Cursor), Pi (Community), Antigravity (Google), and Gemini CLI (Google).

By estimated cost, Claude Code currently leads at 73.5% of total spend, followed by Codex CLI (21.7%), OpenCode (4.4%), Claude Code Desktop (0.2%), OpenCode Go (0.0%), Hermes Agent (0.0%), DeepSeek Harness (0.0%), GitHub Copilot CLI (0.0%), Cursor (0.0%), Pi (0.0%), Antigravity (0.0%), and Gemini CLI (0.0%).

Cost share is influenced by both adoption (number of users) and model pricing (higher-tier models cost more per token). A tool with fewer users but more expensive models can have a larger cost share than a widely-adopted tool using budget models.

clawdboard is the only platform tracking real usage data across all 12 tools side by side, providing a unique view of how developers actually use AI coding assistants in practice.

AI Coding Tool FAQ

Common questions about AI coding tool usage data and comparisons.

[01]What AI coding tools does clawdboard track?

clawdboard tracks usage from 12 AI coding tools: Claude Code (Anthropic), Codex CLI (OpenAI), OpenCode (Community), Claude Code Desktop (Anthropic), OpenCode Go (OpenCode), Hermes Agent (Nous Research), DeepSeek Harness (DeepSeek), GitHub Copilot CLI (GitHub), Cursor (Cursor), Pi (Community), Antigravity (Google), and Gemini CLI (Google). Each tool's usage is tracked separately, allowing side-by-side comparison of cost, token usage, and model preferences.

[02]Which AI coding tool is most popular?

Based on data from 84 developers, Claude Code leads with $933k in estimated API cost (73.5% of total spend). Popularity is measured by estimated cost, which reflects total usage volume weighted by model pricing.

[03]How does clawdboard track usage across different tools?

The clawdboard CLI reads local log files from each supported tool on a developer's machine. Claude Code, Codex CLI, OpenCode, Claude Code Desktop, OpenCode Go, Hermes Agent, DeepSeek Harness, GitHub Copilot CLI, Cursor, Pi, Antigravity, and Gemini CLI all maintain local usage logs. The CLI extracts token counts and model identifiers from these logs and syncs aggregate data to clawdboard. No code, prompts, or conversation content is collected.

[04]Can I use multiple AI coding tools with clawdboard?

Yes. If you use multiple supported tools, the clawdboard CLI will detect and sync usage from all installed tools. Your profile page shows a breakdown by tool, and the leaderboard aggregates your total spend across all tools.

[05]How are costs estimated for different tools?

All costs are estimated from token counts and published API pricing for each model. Usage is priced at the respective provider's rates (Anthropic, OpenAI, etc.). These are API-equivalent estimates — most developers pay flat subscription fees.

[06]How often is tool comparison data updated?

Individual developers sync their usage every 2 hours by default. The aggregate statistics on this page are recalculated hourly. New tools are added to tracking as they gain community adoption.

Track Your AI Coding Usage

Works with Claude Code, Codex CLI, OpenCode, Claude Code Desktop, OpenCode Go, Hermes Agent, DeepSeek Harness, GitHub Copilot CLI, Cursor, Pi, Antigravity, and Gemini CLI. Free, open-source, takes 30 seconds.

$npx clawdboard auth
View LeaderboardAll Statistics