☁️AWS Machine Learning Blog•較早收集於 9m
Claude Opus 4.8 現已於 AWS 上線
💡直接在 AWS Bedrock 上使用 Anthropic 最新的 Claude Opus 4.8，進行生產級的 AI 代理開發。
⚡ 30-Second TL;DR
有什麼變化
Claude Opus 4.8 現已可透過 Amazon Bedrock 使用。
為什麼重要
Opus 4.8 在 AWS 上的可用性，讓企業開發者能在安全且具擴展性的 Amazon Bedrock 生態系統中運用 Anthropic 的最新模型，有助於更輕鬆地部署 AI 代理工作流程。
下一步行動
請檢查 Amazon Bedrock 控制台以啟用 Claude Opus 4.8，並查閱針對代理工作流程的最新整合文件。
誰應關注：Developers & AI Engineers
關鍵要點
- •Claude Opus 4.8 現已可透過 Amazon Bedrock 使用。
- •針對整合至複雜的 AI 代理系統進行了優化。
- •提供針對生產級推論工作負載的具體實作指南。
🧠 深度解析
Web-grounded analysis with 8 cited sources.
🔑 增強重點摘要
- •Claude Opus 4.8 introduces 'dynamic workflows' for Claude Code, allowing the model to plan and execute large-scale, multi-stage problems by running hundreds of parallel subagents in a single session.
- •The model now includes an 'effort control' feature in claude.ai and Cowork, enabling users to specify the computational effort Claude should apply to a response.
- •Opus 4.8's fast mode has been significantly optimized, becoming approximately 2.5 times quicker and costing three times less than its previous iterations.
- •The Messages API for Claude Opus 4.8 now supports system entries within the messages array, providing developers with the flexibility to update the model's instructions mid-task without disrupting the prompt cache or routing.
- •Claude Opus 4.8 demonstrates improved performance across key benchmarks, with agentic coding scores increasing from 64.3% to 69.2%, multidisciplinary reasoning with tools jumping from 54.7% to 57.9%, and knowledge work scores rising from 1753 to 1890.
📊 競品分析▸ Show
| Feature/Model | Claude Opus 4.8 (on Bedrock) | Claude Sonnet 4.6 (on Bedrock) | Llama 3.1 70B (on Bedrock) | Mistral Large 2 (on Bedrock) | GPT-5.2 (General) |
|---|---|---|---|---|---|
| Input Price (per 1M tokens) | $5.00 | $3.00 | $2.65 | $3.00 | ~$15.00 (estimated, based on GPT-5.3-Codex) |
| Output Price (per 1M tokens) | $25.00 | $15.00 | $3.50 | $9.00 | ~$75.00 (estimated, based on GPT-5.3-Codex) |
| Agentic Coding Score | 69.2% (Opus 4.8) | 79.6% (Sonnet 4.6 SWE-bench Verified) | N/A | N/A | ~80.0% (SWE-bench Verified) |
| Multidisciplinary Reasoning with Tools | 57.9% | N/A | N/A | N/A | N/A |
| Context Window | 1M tokens | 200K tokens | N/A | N/A | N/A |
| Availability | AWS Bedrock, Claude Platform on AWS, Google Cloud, Microsoft Foundry | AWS Bedrock, Anthropic API, Google Cloud, Microsoft Foundry | AWS Bedrock | AWS Bedrock | Various platforms (e.g., OpenAI API) |
🛠️ 技術深入
- Enhanced Judgment and Autonomy: Claude Opus 4.8 is designed for sharper judgment, increased honesty about its progress, and the ability to operate independently for longer durations than previous models.
- Reliability and Error Handling: Early testers report that Opus 4.8 is more prone to flag uncertainties and less likely to make unsupported claims, leading to more predictable behavior and fewer review cycles in production.
- Performance Improvements: Significant gains are observed in agentic coding (69.2%), multidisciplinary reasoning with tools (57.9%), agentic computer use (83.4%), and knowledge work (1890 score).
- Dynamic Workflows: A new research preview feature for Claude Code allows the model to undertake larger tasks by planning work and executing hundreds of parallel subagents within a single session, verifying outputs before reporting back.
- Effort Control: Users can now specify the level of effort Claude applies to a response in claude.ai and Cowork, enabling more granular control over output quality and speed.
- Messages API Enhancements: The Messages API now accepts system entries directly within the messages array, facilitating mid-task instruction updates without affecting prompt caching or routing.
- Optimized Fast Mode: The fast mode for Opus 4.8 operates approximately 2.5 times faster and is three times more cost-effective than prior models.
- Context Window: Claude Opus 4.8 supports a 1 million token context window.
- Pricing Structure: On AWS Bedrock, Claude Opus 4.8 is priced at $5.00 per million input tokens and $25.00 per million output tokens, with potential cost savings of up to 90% with prompt caching and 50% with batch processing.
🔮 前景展望AI analysis grounded in cited sources
Increased adoption of autonomous AI agents in enterprise workflows.
Claude Opus 4.8's enhanced agentic capabilities, improved judgment, and ability to handle complex, multi-stage tasks with less oversight will accelerate the development and deployment of more reliable and sophisticated AI agents in production environments.
Further optimization and streamlining of AI development on AWS Bedrock.
The deep integration of Opus 4.8 with AWS Bedrock's managed features like Guardrails and Knowledge Bases, coupled with new features like dynamic workflows and Messages API updates, will simplify and enhance the creation and management of advanced AI applications within the AWS ecosystem.
Intensified competition among frontier LLMs in agentic and coding benchmarks.
The benchmark improvements in Opus 4.8, particularly in agentic coding and multidisciplinary reasoning, will likely prompt other leading AI models to further enhance their own capabilities in these critical areas to maintain competitiveness.
⏳ 時間線
2023-03
Anthropic announces Claude 1
2024-03
Anthropic releases Claude 3 Opus
2025-05
Anthropic releases Claude Opus 4
2025-08
Claude Opus 4.1 becomes available in Amazon Bedrock
2025-11
Claude Opus 4.5 launches on Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry
2026-02
Claude Opus 4.6 released, featuring a 1M token context window in beta
2026-04
Claude Opus 4.7 released and available on Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry
2026-05
Claude Opus 4.8 released and available on Amazon Bedrock and Claude Platform on AWS
📎 來源 (8)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： AWS Machine Learning Blog ↗