⌘K
Change language Switch ThemeSign In
Narrow Mode
Claude 4.6's 1 Million Token Context Window Now Generally Available ===================================================================
Claude 4.6's 1 Million Token Context Window Now Generally Available ===================================================================  ### 宝玉
@dotey
Claude Opus 4.6 和 Sonnet 4.6 的 100 万 token 上下文窗口今天正式全面开放（GA），且不加价，按标准价格计费，无论请求长度是 9K 还是 900K token，单价完全一样。
Opus 4.6 的定价是输入 $5、输出 $25（每百万 token），Sonnet 4.6 是 $3/$15。此前超过 20 万 token 的请求需要加 beta 标头，现在直接生效，已有代码无需改动。
除了价格统一，这次更新还有几个实际变化：单次请求可处理的图片和 PDF 页数从 100 提升到 600；所有上下文长度享受相同的速率限制，不会因为请求变长而被限流。Claude Code 的 Max、Team 和 Enterprise 用户在使用 Opus 4.6 时将默认启用 100 万上下文，意味着对话压缩（compaction）频率大幅降低，更多对话历史得以完整保留。
在准确率方面，Opus 4.6 在 MRCR v2 基准上得分 78.3%，Sonnet 4.6 在 GraphWalks BFS 上得分 68.4%，均为同类前沿模型在百万 token 长度下的最高分。简单说就是：窗口虽大，信息检索能力没有打折。
Anthropic 引用了多家客户的反馈来说明实际效果。
Cognition（Devin 背后的公司）表示，此前大型代码差异塞不进 20 万上下文窗口，需要分批处理，容易丢失跨文件依赖；现在整个差异一次送入，代码审查质量更高，流程也更简单。Ramp 的工程师提到，Claude Code 在调试时经常消耗超过 10 万 token 检索各类数据源，一旦触发压缩就会丢失细节、陷入重复调试循环，百万上下文直接消除了这个问题。
该功能今天起通过 Claude Platform 原生支持，也可通过 Amazon Bedrock、Google Cloud Vertex AI 和 Microsoft Foundry 使用。Show More
#### Claude
@claudeai · 9h ago
1 million context window: Now generally available for Claude Opus 4.6 and Claude Sonnet 4.6.
902
1,344
18.2K
2.9M
Mar 13, 2026, 9:25 PM View on X
6 Replies
7 Retweets
80 Likes
23.2K Views  宝玉 @dotey
One Sentence Summary
Anthropic officially announces the general availability of the 1 million token context window for Claude Opus 4.6 and Sonnet 4.6, maintaining original pricing while significantly enhancing multimodal processing capabilities.
Summary
This tweet details Anthropic's announcement regarding the general availability (GA) of the 1 million token context window for Claude Opus 4.6 and Sonnet 4.6. Key highlights include: 1. Unified pricing, where the unit price remains consistent for both short and long requests (Opus $5/$25, Sonnet $3/$15 per million tokens); 2. Enhanced multimodal capabilities, with the number of supported images and PDF pages per single request increasing from 100 to 600; 3. Optimized rate limits, ensuring long requests are no longer subject to throttling. Performance-wise, Opus 4.6 demonstrates excellent results in long-text retrieval benchmarks. The tweet further illustrates the practical value of the ultra-long context in handling complex code diffs and reducing detail loss from dialogue compaction, citing examples from Cognition and Ramp. The feature is now live on the Claude Platform and concurrently available through major cloud providers like Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
AI Score
88
Influence Score 15
Published At Yesterday
Language
Chinese
Tags
Claude 4.6
Anthropic
Context Window
LLM
AI Model HomeArticlesPodcastsVideosTweets
Claude 4.6's 1 Million Token Context Window Now Generally... ===============