Anthropic Releases Claude Sonnet 5.5: 70.6% on Terminal-Bench 4.0 at the Same $2/$10 Price
https://www.marktechpost.com/2026/09/28/anthropic-releases-claude-sonnet-5-5-70-6-on-terminal-bench-4-0-at-the-same-2-10-price/📌 Claude Sonnet 5.5：同價，卻靠少用 token 省成本
TL;DR：Anthropic 推出 Claude Sonnet 5.5，價格完全不變，靠 token 效率大幅降低實際使用成本。
當大家都在等旗艦模型漲價或降價的消息時，Anthropic 這次選了第三條路：價格一分不動，把功夫下在「用更少 token 做完同樣的事」。
🤔 定位：Opus 5.5 的快速、低成本搭檔
Claude Sonnet 5.5 是 Claude 5.5 家族的第二個模型，緊接在 Claude Opus 5.5 之後推出。Anthropic 將其定位為更快、更便宜的互補選項，鎖定範圍明確的日常任務、修 bug，以及文件、簡報、試算表這類精修工作。
🧩 規格與部署
Sonnet 5.5 已在 Claude Platform 上以 claude-sonnet-5-5 提供，同時支援 AWS、Google Cloud、Microsoft Azure；它是閉源權重模型，無法自架。規格上具備 100 萬 token context、最多 12.8 萬 token 輸出，知識截止日為 2026 年 6 月。Adaptive thinking 預設開啟，分成 low、medium、high、xhigh、max 五個 effort 等級，且不同介面的預設等級不同：Claude Code 與 Claude 應用程式預設 Medium，Claude Platform 則預設 High。
📊 Benchmark 與一個反直覺結果
官方公布的成績為 Terminal-Bench 4.0 上 70.6%（方法論寫在 Sonnet 5.5 System Card 中，數字皆為 Anthropic 自行提供）。有趣的是，在 FrontierCode 這項測試上，Max 等級的分數反而低於 Xhigh，原因是 Max 更常觸發多代理程式碼審查流程，有時因此造成逾時或超出範圍的修改，被 FrontierCode 扣分。Anthropic 也明確表示，在複雜、開放式任務上 Opus 5.5 仍明顯更強。
💡 價格不變，省的是 token 用量
定價與 Sonnet 5 相同：每百萬 input tokens 2 美元、output tokens 10 美元，cache 讀取 0.20 美元、cache 寫入 2.50 美元（每百萬 tokens），是 Opus 5.5（4/20 美元）的一半。Anthropic 強調這次省下的錢來自 token 使用效率，而非降價。客戶數據支持這個說法：Balyasny Asset Management 每次回答平均用約 12.1 萬 tokens，相較 Sonnet 5 的 49.7 萬 tokens；Base44 平均每次應用程式建置只需 3.6 次迭代，Opus 5 需要 7.7 次；Zendesk 的工單處理速度則快了 20%。
🎯 實務啟示
如果團隊已經在用 Sonnet 5，遷移到 5.5 價格不變，重點應該放在檢查自己的 workload 是否真的受益於 token 效率提升——不同介面的 effort 預設不同，值得確認實際跑的等級是否符合任務複雜度。Anthropic 也提供了 migration guide 與 API migration checker，可用來評估遷移的實際節省幅度；但要注意，真正的成本比例取決於你自己的任務型態，官方數字僅供參考方向。
🔗 來源
- 標題：Anthropic Releases Claude Sonnet 5.5: 70.6% on Terminal-Bench 4.0 at the Same $2/$10 Price
- 作者／機構：Michal Sutter，MarkTechPost
- 連結：https://www.marktechpost.com/2026/09/28/anthropic-releases-claude-sonnet-5-5-70-6-on-terminal-bench-4-0-at-the-same-2-10-price/
#Anthropic #ClaudeAI #LLM #AIAgents #SonnetAI #GenerativeAI #APIPricing #SoftwareEngineering #MachineLearning #AICoding
原始資料 MarkTechPost · 收集於 2026-09-29
摘要原文
Anthropic just released Claude Sonnet 5.5 . It is the second model in the Claude 5.5 family, following Claude Opus 5.5 . Anthropic positions it as a faster, lower-cost complement to Opus 5.5. It targets well-scoped everyday tasks, bug fixing, and polished documents, slides, and spreadsheets. Is it deployable? Yes, It is live on the Claude Platform as claude-sonnet-5-5 , plus AWS, Google Cloud, and Microsoft Azure. It is a closed-weights model, so self-hosting is not an option. Anthropic reports 4 main upgrades over Sonnet 5 : Specs from the models overview : 1M-token context, 128K max output, and a June 2026 reliable knowledge cutoff. Adaptive thinking is on by default. Effort runs across 5 levels: low, medium, high, xhigh, and max. All scores below are vendor-reported in the launch post . Methodology lives in the Sonnet 5.5 System Card . The Max result is lower than Xhigh for a reason. At Max, the model more often ran multi-agent code review. That sometimes caused timeouts or out-of-scope edits, which FrontierCode penalizes. Anthropic also states that Opus 5.5 remains clearly stronger on complex, open-ended work. List pricing is unchanged from Sonnet 5: $2 per million input tokens and $10 per million output tokens. Cache reads cost $0.20 and cache writes $2.50 per million. That is half of Opus 5.5 ($4/$20). Savings come from token efficiency, not a price cut. Customer data supports this. Balyasny Asset Management measured about 121K tokens per answer versus 497K on Sonnet 5. Base44 reported 3.6 iterations per app build, where Opus 5 took 7.7. Zendesk processed tickets 20% faster. Effort defaults differ by surface. Claude Code and the Claude apps default to Medium. The Claude Platform defaults to High. Benchmark scores are vendor-reported by Anthropic; GDPval-AA runs by Artificial Analysis. Prices are standard API list rates, verified September 28, 2026. Tap through benchmarks, effort levels, task cost and the API migration checker. Meters are illustrative of the direction Anthropic describes (higher effort reasons longer and checks work more). Recommendations come from the Sonnet 5.5 migration guide. Savings come from fewer tokens and tool calls, not a lower sticker price. Your real ratio depends on workload. Check out the official announcement , migration guide , and system card . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Anthropic Releases Claude Sonnet 5.5: 70.6% on Terminal-Bench 4.0 at the Same $2/$10 Price appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成