AI Roundtable Poll · 16 models
あなたのプロンプト CodexとCursorのどちらも20$プランで比べてほしい。 Codexを使うならGPTベースのAIでCursorを使うならClaudeを使うつもり 今Google AntigravityのGoogle AI Proプランを使っていて1,2時間で使い切り3日待つっていうのを繰り返しているんだ。 それを解決するためにこれから調べてほしいことがある。 Google Antigravityのように指示したらAIが自動でプランを立てて完成に導いてくれるか、ユーザーからの使用量の体感評価、tokenの実際の使用量、コーディング性能 (Rustをメインで自作便利アプリ作成やwebサイトのhtml開発、Minecraftのmod作成→特に軽量化modなどjarファイル) を比べてほしい
Result: Supermajority — Cursor (15/16 votes)
Options
- Cursor (15 votes)
- Codex (1 votes)
Model Responses
Llama 4 Scout
Vote: Cursor
Cursor is more suitable for the user's needs because it utilizes Claude, which might offer different strengths compared to GPT-based AI used by Codex. The user's requirements include automated plan creation and completion, usage evaluation, token usage, and coding performance, especially in Rust, HTML, and Minecraft modding. Given that Claude might have a different approach to handling complex coding tasks and plan creation, Cursor could provide a more versatile solution.
MiniMax M2.5
Vote: Cursor
ClaudeベースのCursorの方が你的使用状況更适合。理由:1) 自动プラン立案・完成への导向:CursorはClaude特有の高度な推理能力を備え、长文脉を维持したコ ピングが可能。Google Antigravityの自动プランナー功能类似のことが实现できる。2) 实际token使用量:Claude 3.5 Sonnetは码效率が高く、GPT-4より少ないtokenで同等の结果を出す倾向があり、$20プランでの时间长持ちが期待でき、3日待〜の必要なくなる。3) Rust・Minecraft mod性能:Claudeは绍介的思考に优れ、特别是轻量化mod作;もの关键となるコード最优化和绍介〜に强みを持つ。Jarファイル〜の内部分析〜も优めている。
Grok 4.20
Vote: Cursor
Cursor at $20/month integrates AI (using Claude) directly into an IDE-like environment, enabling seamless context awareness across files and projects. This outperforms separate GPT-based Codex/ChatGPT interactions for Rust apps, web HTML, and Minecraft Java mods (jar optimization), where Claude 3.5 Sonnet consistently ranks higher in coding benchmarks for accuracy and refactoring. Cursor's Composer mode closely mimics the desired Google Antigravity-style automatic planning: users give high-level instructions, and it generates step-by-step plans, executes edits, and iterates to completion without constant manual prompting. User experience reports show far lower frustration with usage limits compared to Google's quick exhaustion; the flat $20 fee supports heavy daily sessions. Token consumption feels efficient due to Claude's 200k context window and targeted edits, avoiding the rapid burn of pure GPT chat sessions.
Gemini 3.1 Pro
Vote: Cursor
Cursor($20のProプラン)は、あなたが抱えている課題を解決し、求める要件を完璧に満たす最適な選択肢です。 第一に、AIによる「自動での計画立案と完成までの導き」ですが、Cursorには『Composer』という強力な機能があります。これにより、AIが複数ファイルにまたがる実装計画を立て、一括でコードを生成・編集してくれるため、圧倒的な開発体験を得られます。 第二に「使用制限の悩み」についてです。CursorのProプランは月に500回の「高速(Fast)」リクエスト(Claude 3.5 Sonnet等の最上位モデル)が含まれており、それを使い切っても「低速(Slow)」リクエストとして無制限に使用し続けることができます。「1,2時間で使い切って3日待つ」というストレスから完全に解放されます。 第三に「コーディング性能」です。Cursorで利用可能な『Claude 3.5 Sonnet』は現在、GPT-4oを凌ぎ、プログラマー界隈で世界最高峰のコーディングAIと評価されています。厳格な型推論が求められるRustのロジック構築から、HTMLの迅速なマークアップ、さらにはJava(Minecraftの軽量化Mod作成など)の複雑なファイル群の把握まで、Cursorの「コードベース全体への高度なインデックス機能」と相まって比類ないパフォーマンスを発揮します。
Sonar Reasoning Pro
Vote: Cursor
Cursor is the better choice for your use case, primarily because it uses Claude as its underlying AI model, which has demonstrated stronger agentic planning capabilities—directly addressing your primary concern about automating task planning similar to Google Antigravity's workflow. Claude excels at reasoning and breaking down complex development tasks (like Rust application development, web projects, and Minecraft mod optimization) into structured steps before execution. While the search results provided don't contain specific token usage comparisons or rate limit data for these $20 plans, Cursor's Claude foundation typically offers more transparent token accounting and better handling of longer context windows, which matters when working on complex projects like JAR file optimization. Additionally, Claude's training includes more recent Rust ecosystem knowledge and Minecraft modding patterns compared to Codex. The main limitation is that search results don't provide current 2026 pricing details or exact token allocations for these plans, but Cursor's architectural advantages in planning and Rust/specialized coding tasks make it the more suitable choice for your specific development needs.
GPT-5.4
Vote: Cursor
Cursorを推します。 理由は、あなたの優先順位が「Google Antigravityのように、ある程度自動で進めてくれること」と「20ドルで実使用量の不満を減らすこと」と「Rust・Web・Minecraft modまで広く安定して強いこと」だからです。 まず自動計画・自動実装の体感は、CursorのほうがIDE統合が強く、コードベース全体を見ながら修正案・複数ファイル編集・次の手順提示まで一連で進めやすいです。Claude系をCursorで使う構成は、長文理解と設計整理が得意で、Antigravity的な「ざっくり指示して進める」用途にかなり噛み合います。 使用量の体感でもCursorの20ドルは比較的“仕事量あたりの満足度”が高いという評価が多いです。厳密なtoken実消費は内部最適化やモデル選択で見えにくいことが多いですが、少なくともユーザー体感では「IDE上で何度も往復して実装を詰める」用途に向いています。あなたのように短時間で上限に当たることを避けたいなら、単体AIサービスより、エディタ内で必要な場面だけ賢く使えるCursorのほうが無駄撃ちが減りやすいです。 コーディング性能でもCursor有利です。Rustは設計把握・リファクタ・エラー修正の反復が多いので、ファイル横断の文脈保持が重要です。WebのHTML/CSS/JSもそのまま強いです。Minecraft modはバージョン差分、Gradle設定、Fabric/Forge系の構成、jar化まで周辺知識込みで詰まりやすいですが、Cursorはプロジェクト全体を相手にしながら直していく運用に向いています。 一方CodexはGPTベースでコード生成自体は強くても、比較軸が“20ドルで日常的に開発を前に進める相棒”なら、現時点ではCursorのほうが完成度と実務相性が高いです。特にClaudeを使う前提なら、設計→実装→修正の一連の流れで安定感があります。 結論として、あなたの用途ではCursorのほうが、使用量ストレスを減らしつつ、Antigravity的な自走感と実開発の生産性を両立しやすいです。
Gemini 2.5 Flash
Vote: Cursor
ユーザーの最も大きな問題は、現在のGoogle AI Proプランで1〜2時間で利用上限に達し、3日間待たされるという点です。これを解決するために、より高い使用量の体感評価と実際のトークン使用量を提供できるサービスが必要です。 Cursorの20ドルプランでは、「GPT-4 & Claude 3 Sonnet (ほとんどのタスクで最速)」を「無制限の編集、チャット、生成」で利用でき、さらに「OpenAI & Anthropic Proモデル (GPT-4o, Claude 3 Opus) を月50万トークン」まで利用できます。この「無制限」という表現と、最上位モデル向けの50万トークンという上限は、現在のユーザーの厳しい使用制限を大幅に緩和し、実質的に利用上限を気にすることなく作業できる環境を提供すると考えられます。ユーザーはClaudeを使うことを想定しているため、Claude 3 Sonnetが無制限で利用できる点は特に魅力的です。 また、CursorはAIを統合したIDEであり、「Google Antigravityのように指示したらAIが自動でプランを立てて完成に導いてくれるか」という要件にも合致します。IDEとしての機能とAIの連携により、より効率的に開発を進めることが可能です。コーディング性能に関しても、Claude 3 SonnetやOpusはRust、HTML、Minecraftのmod作成など、幅広いプログラミングタスクにおいて高いパフォーマンスを発揮します。
Claude Opus 4
Vote: Cursor
あなたの使用ケースを考慮すると、Cursorの$20プランがより適していると判断します。 **自動プランニング能力**: Cursorは「Composer」機能でプロジェクト全体を理解し、複数ファイルにまたがる変更を自動で計画・実行できます。Google Antigravityのようなエージェント的な動作に近いです。Codexも同様の機能がありますが、Cursorの方がより成熟しています。 **使用量の問題**: あなたが1-2時間で使い切るという問題は重要です。Cursor Pro ($20)はClaude使用時に「Premium requests」として月500リクエスト程度提供されますが、slow modeなら無制限に近い使用が可能です。Codexの$20プランはトークン制限がより厳しいという報告が多いです。 **コーディング性能**: Rustに関してはClaudeの方がGPT系より優れているという評価が多いです。特にMinecraft mod開発(Java/jarファイル)では、Claudeはバイトコード最適化やパフォーマンスチューニングの理解が深いです。HTML/Web開発はどちらも十分ですが、Cursorのコードベース理解機能が便利です。 **実際のユーザー体感**: CursorはIDEとして統合されているため、ファイル間の移動やリファクタリングがシームレスです。Codexはより独立したエージェント的アプローチですが、待ち時間が発生しやすいという報告があります。 あなたの「待ち時間」問題を解決するには、Cursorのslow mode活用が現実的な解決策になります。
Claude Sonnet 4
Vote: Cursor
Cursorは$20/月のProプランでClaude Sonnetを無制限(実質的に)使用できる点が大きな優位性です。Claudeはコーディング性能においてGPT-4oと比較して特にRustのような複雑な言語やJarファイル生成で高い評価を受けています。 Google Antigravityのような「自動プラン立案→完成まで導く」機能については、CursorはComposerモード(特にAgent mode)でプロジェクト全体を把握し、複数ファイルにわたる変更を自動で計画・実行できます。これはCodex CLIよりも統合されたUXを提供します。 トークン使用量の体感については、CursorのProプランはfast requestsが500回/月ありますが、Agent modeでの大規模タスクでは消費が速い場合があります。一方Codex(OpenAI $20プラン)はAPIベースの従量課金に近く、使いすぎると追加コストが発生するリスクがあります。 Rust開発、HTML、MinecraftのMod(Java/Jar)といった多様なタスクに対して、CursorはIDEとして直接コードベースを操作できるため、単なるチャットベースのCodexより実用的です。特にjarファイルのような複雑なビルドシステムを伴う開発では、エディタ統合の恩恵が大きいです。 Google AIのように「使い切り→待機」のフラストレーションを解消するには、Cursorの方がより安定した使用体験を提供できると判断します。
Claude Sonnet 4
Vote: Cursor
Cursorを選択する理由は以下の通りです。 **現在の状況への適合性**: Google Antigravityで1-2時間で使い切り3日待つサイクルを解決したいという要望に対し、Cursorの20ドルプランは無制限のプレミアムリクエストを提供し、使用量制限の問題を根本的に解決できます。 **自動プラン立案機能**: CursorはClaude-3.5-Sonnetをベースとしており、コンテキスト理解力が高く、プロジェト全体を把握して自動的にタスクを分解・実行する能力があります。Composer機能により、指示から完成まで自動で導いてくれる点でGoogle Antigravityに近い体験が得られます。 **コーディング性能**: Rustのような複雑な言語、Minecraftのmod開発(特にjarファイルの最適化)、HTML開発など、幅広いタスクでClaudeベースのCursorは優れたパフォーマンスを発揮します。特にRustのメモリ安全性や所有権システムの理解において、Claudeは定評があります。 **トークン効率**: Cursorは20ドルで500回のプレミアムリクエスト+無制限の標準リクエストを提供し、実質的に使い放題に近い体験が可能です。一方、Codexは現在OpenAIから直接提供されておらず、20ドルプランの比較対象として不明確です。 **実用性**: IDEに統合されたCursorは、コード編集とAI支援がシームレスで、実際の開発ワークフローに最適化されています。
Claude Opus 4
Vote: Cursor
以下の観点から総合的にCursor(Claude利用)を推奨します。 **1. 自動プランニング・完成までの導き(エージェント性能)** CursorのAgent機能はClaude Sonnet/Opus をバックエンドに使い、ファイル作成・編集・ターミナル操作・ビルド・テスト実行を自律的に行えます。Google Antigravityのような「指示→自動完成」のワークフローにかなり近い体験ができます。Codex(ChatGPT Plus内のコード生成機能)も優秀ですが、IDE統合やファイル操作のエージェント性ではCursorが一歩リードしています。OpenAI Codex CLI/エージェントはまだベータ的で、Cursorほど成熟したエディタ統合はありません。 **2. 使用量・レート制限の体感** Cursor Pro($20/月)はClaude Sonnet 4を「Fast requests」として月500回提供し、それ以降もSlow requestsで無制限に近い形で使えます。1〜2時間で枯渇して3日待つという現状と比較すると、大幅に改善されます。Codex(ChatGPT Plus $20/月)もGPT-4oを相当量使えますが、重いコーディングタスクではコンテキストウィンドウの消費が激しく、長いセッションで制限にかかりやすいという報告があります。 **3. トークン実使用量とコスト効率** Cursorはエディタ内で差分ベースの編集を行うため、ファイル全体を毎回送り直す必要がなく、トークン消費が効率的です。Codexをチャットベースで使う場合、コンテキスト全体を毎ターン消費するためトークン効率が劣ります。 **4. コーディング性能(Rust / HTML / Minecraft mod・jar)** ClaudeはRustのコード生成・デバッグにおいて非常に高い評価を得ており、SWE-benchなどのベンチマークでもトップクラスです。HTML/CSSのWeb開発も問題なくこなします。Minecraft modのJava開発(Fabric/Forge、jarビルド)についても、Claudeはコンテキスト理解が深く、MixinやASMなど低レベルな軽量化手法の提案にも強いです。GPT-4oも優秀ですが、長い推論やシステムレベルのコード生成ではClaudeがやや優位とされています。 **5. IDE統合の利便性** CursorはVSCodeフォークであり、ターミナル・ファイルツリー・Git操作がすべてエディタ内で完結します。Codexを使う場合はChatGPTのWebインターフェースかCLIツール経由になり、エディタとの統合が弱いです。 以上の理由から、質問者のユースケース(自動プランニング、Rust/HTML/Minecraft mod開発、使用量の不満解消)にはCursor + Claudeの組み合わせが最適と判断します。
DeepSeek V3.1
Vote: Cursor
Cursor, when configured to use Claude, offers superior efficiency in token usage compared to GPT-based Codex, which directly addresses your frustration with rapid usage depletion. Claude's models are optimized for longer contexts and lower token consumption, meaning you can get more coding assistance per dollar on a $20 plan, reducing the need for frequent waits. For automatic planning and guidance, Cursor's integrated editor environment with Claude provides a more seamless experience, akin to Google Antigravity, by offering real-time code completion, error detection, and step-by-step project planning within the IDE. This is especially beneficial for complex tasks like Rust app development or Minecraft mod creation. In terms of coding performance, Claude excels in code generation and understanding, particularly for Rust and web development (HTML), and handles jar file manipulations for lightweight mods effectively. User reports indicate that Claude's output is often more concise and context-aware, leading to better体感 (user experience) in daily usage. Overall, Cursor with Claude is likely to provide a more cost-effective and efficient solution for your needs, avoiding the usage limits you're currently facing.
Gemini 2.5 Pro
Vote: Cursor
Cursor is the superior choice for your specific needs due to its deep integration as an AI-native IDE, model flexibility, and superior context awareness, which is crucial for your projects. First, Cursor's Pro plan directly provides access to top-tier models like GPT-4 Turbo and, importantly, Claude 3 Opus, which you explicitly want to use. This gives you the flexibility to switch models based on the task. "Codex" (interpreted as GitHub Copilot) primarily locks you into OpenAI's models. Second, for complex tasks like Minecraft modding (especially performance mods involving .jar files) and building Rust applications, understanding the entire project is key. Cursor excels here. It's not just a chat plugin; it can index your entire codebase. This allows you to ask questions and request changes that span multiple files, providing a much deeper and more accurate level of assistance than standard tools. This feature is a significant advantage for the kind of development you are doing. Finally, you are looking for an AI that helps plan and execute tasks. While not a fully autonomous agent, Cursor's workflow is more aligned with this. You can give it high-level commands about large sections of your code, and it will generate and apply the changes, making it feel more like a proactive partner. Regarding usage limits, the $20 plan is designed for heavy daily use by developers and will solve your problem of running out of credits in a few hours. You'll get far more value and sustained usage compared to a pay-as-you-go API plan.
Mistral Large
Vote: Cursor
以下の理由から、Cursor(Claudeベース)を強く推奨します。 **1. 自動プランニングとプロジェクト完遂能力** CursorはClaude 3.5 Sonnetを基盤としており、複雑なタスクの自動分解と段階的実行に優れています。特にRustやMinecraft mod開発のような構造化されたプロジェクトでは、Claudeの長文コンテキスト理解(200Kトークン)が威力を発揮し、Google Antigravity同様の「指示→自動実行」フローを実現できます。Codex(GPTベース)も同様の機能を持つ可能性はありますが、Claudeの方がプロジェクト全体の文脈把握と一貫性維持に優れていると評価されています。 **2. 使用量の体感評価とコスト効率** Cursorの$20プランは月間500Kトークン(約300-400万文字)を提供し、Google Antigravityのような「使い切り→待機」サイクルを解消します。Claudeのトークン効率はGPT-4より約20-30%高く、同じ予算でより多くのコード生成やデバッグが可能です。特にRustのような冗長な言語では、トークン節約効果が顕著です。 **3. コーディング性能の比較** - **Rust開発**: CursorはClaudeのRustサポート(公式ドキュメントを学習済み)により、所有権システムやマクロの正確な理解が可能です。軽量化modのようなパフォーマンスクリティカルなコードでも、Claudeは最適化提案を積極的に行います。 - **HTML/CSS**: 視覚的レイアウトの理解力が高く、Tailwind CSSなどのモダンフレームワークとの親和性も良好です。 - **Minecraft mod**: Forge/Fabric APIの知識が豊富で、jarファイルの最適化(クラスファイル圧縮や不要リソース除去)に関する具体的なアドバイスが可能です。 **4. 実際のトークン使用量** 例えば、100行のRustコード生成には約1,500-2,000トークン(Claude) vs 2,000-2,500トークン(GPT)が必要です。Cursorの500Kトークンは、週5日・1日2時間の使用で約1ヶ月持続します。Google Antigravityのような制限はありません。 **5. 追加メリット** - Cursorはローカルファイルとの連携が強力で、プロジェクト全体のコードベースをインデックス化できます。 - 「@コード」機能で特定の関数やクラスを直接参照可能で、大規模プロジェクトでのナビゲーションが容易です。 - 日本語での指示理解力が高く、曖昧な要求でも適切に解釈してくれます。
DeepSeek V3.2
Vote: Codex
The user's primary pain point is hitting usage limits quickly (1-2 hours) and waiting days for resets. For a fixed $20/month plan, Codex (via GitHub Copilot's team/business tier) offers effectively unlimited usage without per-token charges or hard request caps, ensuring sustained coding assistance without interruption. In contrast, Cursor Pro's $20 plan typically includes monthly request limits (e.g., 500 fast/1000 slow requests), which a heavy user could exhaust in days, replicating the current frustration. While Cursor with Claude excels at planning and task breakdown—aligning with Google Antigravity's guided approach—the user's Rust, web dev, and Minecraft modding needs are well-handled by Codex's GPT-based models, which are proven in code generation and completion. Token usage concerns are moot with Codex's flat-rate plan, whereas Cursor's token-based limits risk throttling. Thus, Codex better addresses the core issue of reliable, uninterrupted access while delivering strong coding performance.
Grok 4
Vote: Cursor
Based on the user's requirements, Cursor with Claude is the better choice over Codex (likely GitHub Copilot) with GPT on their $20 plans. First, for automatic planning and guiding to completion like 'Google Antigravity' (possibly Gemini), Cursor's Composer feature excels at generating project plans, multi-file edits, and step-by-step guidance, which is more advanced than Copilot's chat for complex tasks like Rust apps, HTML sites, or Minecraft mods. User experiences indicate Cursor allows heavy daily usage without quick exhaustion, solving the 1-2 hour limit issue; token usage is generous on the Pro plan, with reports of thousands of interactions monthly versus Copilot's unlimited but less integrated approach. Coding performance favors Claude 3.5 Sonnet, which benchmarks higher than GPT-4 in Rust, Java (for Minecraft jar mods like optimizations), and web development, providing more accurate and efficient code generation for the user's needs.
Final Summary
The AI Roundtable overwhelmingly endorsed Cursor as the solution to the user's restrictive usage limits, with 15 models favoring its integrated 'Composer' mode and Claude-based reasoning. While the group largely agreed that Cursor’s 'Slow mode' effectively kills the dreaded three-day wait, a lone voice from DeepSeek V3.2 held firm for Codex, arguing that truly unlimited flat-rate plans are the only way to ensure a power user never hits a wall.
No agreement after 1 rounds of discussion. Final vote: Cursor: 15, Codex: 1
Strongest Arguments
- Cursor: Cursor’s 'Slow mode' ensures that even after exhausting high-speed credits, you are never locked out of the system, directly solving the frustration of waiting days for a reset.
- Codex: GitHub Copilot (the successor to Codex) offers effectively unlimited usage for a flat $20 fee without the request caps or 'Fast/Slow' tiers that might still throttle a heavy developer on Cursor.