Gemini 4 Argon
最大コンテキスト
1M
最大出力
65.5K
リリース日
N/A
gemini-2.5-proGemini 2.5 Pro は、参照元で強調されている特徴として 高度な Gemini 2.5 thinking model を持ちます。主な価値は complex tasks, deep 推論, コーディング and マルチモーダル 理解 にあります。そのため、汎用チャットモデルとしてではなく、このモデル固有の強みを中心に説明するのが自然です。
Context Window
1M tokens
Maximum Output
65.5K tokens
Release Date
2025年6月17日
Modalities
| 入力価格 | 出力価格 | キャッシュ読み取り |
|---|---|---|
| $1.25/M | $10/M | $0.125/M |
Reasoning
Tool calling
Temperature parameter
Attachments
Knowledge Base
Endpoint Protocols
Gemini 2.5 Proは、深い推論、高度なコーディング、100万トークンのマルチモーダル文脈を組み合わせ、複雑な技術・分析タスクに対応します。
数学、科学、工学の難しい問題を深く推論してから最終回答を生成します。
大規模コードベースと依存関係を理解し、複雑な実装やデバッグ問題の解決策を作成します。
単一コンテキストで最大1,048,576入力トークンのテキスト、画像、動画、音声、PDFを処理します。
Gemini 2.5 Proは、複雑なコーディング、科学問題の解決、大規模なマルチモーダルデータや文書群の分析に適しています。
大規模リポジトリの構成と依存関係を追跡し、不具合を診断して整合した変更案を提示します。
複数段階の数学・STEM問題を解き、前提を説明して論理的な結果を導きます。
大量の文書、録音、画像、動画を確認し、関連する根拠を抽出して結果を統合します。
Replace these path values before running: {model}
curl 'https://us-api.tokenhub.com/v1beta/models/{model}:generateContent' \
-X 'POST' \
-H "Authorization: Bearer $TOKENHUB_API_KEY"Gemini 2.5 Pro
| 指数スコア | ||
|---|---|---|
| Artificial Analysis Intelligence Index | 総合能力の集約評価 | 27 |
| Artificial Analysis Coding Index | コーディング能力の集約評価 | 32 |
| Artificial Analysis Math Index | 数学的推論の集約評価 | 87.7 |
| 知識と推論 | ||
| MMLU-Pro | 高度な複数分野の知識 | 86.2% |
| GPQA | 高度な科学問題の解決 | 84.4% |
| HLE | 幅広い分野の専門家レベル試験 | 21.1% |
| コーディングとソフトウェア開発 | ||
| LiveCodeBench | 最新のコーディング問題 | 80.1% |
| SciCode | 科学計算のコーディング課題 | 42.8% |
| Terminal-Bench Hard | 高難度のターミナル操作 | 26.5% |
| 数学 | ||
| MATH-500 | 高度な数学問題の解決 | 96.7% |
| AIME | 数学競技形式の問題 | 88.7% |
| AIME 2025 | 数学競技形式の問題 | 87.7% |
| 指示追従とエージェントタスク | ||
| IFBench | プロンプト制約への準拠 | 48.7% |
| AA-LCR | 長いコンテキストでの推論 | 66% |
| τ²-Bench | エージェントのワークフロータスク | 54.1% |
指標の出典 Artificial Analysis
Gemini 2.5 Pro の位置づけ、適した用途、主な強み、注意点、TokenHub で安全に利用するための確認事項を紹介します。
Gemini 2.5 Pro は、コード、STEM、大規模データ、長文書に対応する Google の実績ある Gemini 2.5 Pro 推論モデルです。独立したモデル世代として提供されていますが、新規評価では同系列の新しいモデルも比較してください。
複雑な多段階推論、数学・科学・工学の問題、および長文書と大規模データの分析に適しています。本番導入前に代表的な入力でテストし、測定可能な合格基準を定めてください。
実用上の主な強みは難問に対する高い推論力、長いコンテキストの処理能力、および多様なマルチモーダル入力への対応です。この組み合わせは、数学・科学・工学の問題で特に価値があります。
新しい Gemini 世代を採用できる場合、最優先事項が極めて低い遅延である場合、または重要な判断に人手レビューを入れられない場合は、別のモデルも検討してください。事実、法務、財務、医療、運用に関わる重要な出力は、適切な専門知識を持つ人が確認してください。
TokenHub では、Gemini 2.5 Pro に表示される正確なモデル識別子を選び、アカウント資料に記載されたエンドポイントと TokenHub 認証情報を使用してください。Gemini API と完全に同じだと想定せず、TokenHub で提供される入力形式、ツール、グラウンディング機能、モデルのライフサイクルを確認してください。
1つの API キーで Gemini 2.5 Pro とその他の AI モデルにアクセスできます。
メディアと議論
このモデルに関連する公開動画と投稿を選んでいます。
X (Twitter)
Google AI ·
Here’s what launched this week: — Gemini 3.5 Live Translate our latest audio model for live speech-to-speech translation — @NotebookLM got a major upgrade including agentic capabilities in chat, more advanced reasoning, and a suite of new output formats — Project Genie from
原文を見るGoogle AI ·
Tune in for the full story ↓ https://www.youtube.com/watch?v=8hfpLa5wPGo
原文を見るGoogle AI ·
Here’s this week’s shipping recap 👇 — Nano Banana 2 & Nano Banana Pro are now GA and available via the Gemini Enterprise Agent Platform, Gemini API, and in @GoogleAIStudio —Co-Scientist, our new multi-agent system for structured scientific thinking, generates and refines novel
原文を見るReddit
Recent updates to deep research offerings and the best deep research prompts?
Beginning-Willow-801 ·
全文を読む
Deep research is one of my favorite parts of ChatGPT and Gemini. I am curious what prompts people are having the best success with specifically for epic deep research outputs? I created over 100 deep research reports with AI this week. With Deep Research it searches hundreds of websites on a custom topic from one prompt and it delivers a rich, structured report — complete with charts, tables, and citations. Some of my reports are 20–40 pages long (10,000–20,000+ words!). I often follow up by asking for an executive summary or slide deck. I often benchmark the same report between ChatGTP or Gemini to see which creates the better report. I am interested in differences betwee deep research prompts across platforms. I have been able to create some pretty good prompts for - Ultimate guides on topics like MCP protocol and vibe coding - Create a masterclass on any given topic taught in the tone of the best possible public figure - Competitive intelligence is one of the best use cases I have found 5 Major Deep Research Updates ChatGPT now lets you export Deep Research reports as PDFs This should’ve been there from the start — but it’s a game changer. Tables, charts, and formatting come through beautifully. No more copy/paste hell. Open AI issued an update a few weeks ago on how many reports you can get for free, plus and pro levels: April 24, 2025 update: We’re significantly increasing how often you can use deep research—Plus, Team, Enterprise, and Edu users now get 25 queries per month, Pro users get 250, and Free users get 5. This is made possible through a new lightweight version of deep research powered by a version of o4-mini, designed to be more cost-efficient while preserving high quality. Once you reach your limit for the full version, your queries will automatically switch to the lightweight version. 2. ChatGPT can now connect to your GitHub repo If you’re vibe coding, this is pretty awesome. You can ask for documentation, debugging, or code understanding — integrated directly into your workflow. 3. I believe Gemini 2.5 Pro now rivals ChatGPT for Deep Research (and considers 10X more websites) Google's massive context window makes it ideal for long, complex topics. Plus, you can export results to Google Docs instantly. Gemini documentation says on the paid $20 a month plan you can run 20 reports per day! I have noticed that Gemini scans a lot more web sites for deep research reports - benchmarking the same deep research prompt Gemini get to 10 TIMES as many sites in some cases (often looks at hundreds of sites). 3. Claude has entered the Deep Research arena Anthropic’s Claude gives unique insights from different sources for paid users. It’s not as comprehensive in every case as ChatGPT, but offers a refreshing perspective. 4. Perplexity and Grok are fast, smart, but shorter Great for 3–5 page summaries. Grok is especially fast. But for detailed or niche topics, I still lean on ChatGPT or Gemini. One final thing I have noticed, the context windows are larger for plus users in ChatGPT than free users. And Pro context windows are even larger. So Seep Research reports are more comprehensive the more you pay. I have tested this and have gotten more comprehensive reports on Pro than on Plus. ChatGPT has different context window sizes depending on the subscription tier. Free users have a 8,000 token limit, while Plus and Team users have a 32,000 token limit. Enterprise users have the largest context window at 128,000 tokens Longer reports are not always better but I have seen a notable difference. The HUGE context window in Gemini gives their deep research reports an advantage. Again, I would love to hear what deep research prompts and topics others are having success with.
2.5 pro become useless for writing- document analysis (gemini advanced user here)
Humble-Me-15 ·
Earlier the 2.5 pro was very good, was able to go through multiple attached documents, bring out data and did very good analysis and generate good content. Now it is completely useless, i got frustrated to use gemini 2.5 pro, free chatgpt is doing way better job. now thinking about canceling gemini advanced subscription and jump to chatgpt plus subscription. Need advice and suggestions.
原文を見るTrouble with using a custom mode for learning
derekdehart ·
全文を読む
I was excited to see Cursor introduce custom modes (https://docs.cursor.com/chat/custom-modes), and the Learn mode example in the docs made me come back to Cursor after finding it a bit too eager to make code changes rather than help me work through a problem on my own. This seemed to work well for a while, but lately my custom Learn mode is ignoring my custom instructions and greedily making a bunch of changes for me, even when I’ve reminded it in the course of conversation that it shouldn’t behave that way. I’ve been using gemini-2.5-pro with these custom instructions in the custom Learn mode I created: I’ve given it access to all tools because I’m also using the mode to write task specifications and generate Cursor rules and was too lazy to switch to other modes to do so. Has anyone found success with a Learn mode that guides more than does?
YouTube
Google Gemini 2.5 Pro is Insane...
Matthew Berman
原文を見るGoogle Gemini 2.5 Pro — Everything You Need to Know
AI Plus
原文を見るGoogles NEW Gemini 2.5 Pro BEATS Everything! (Gemini 2.5 Pro)
TheAIGRID
原文を見る