Google Releases Gemini 4 Argon API With 1M Token Output Limit
https://www.marktechpost.com/2026/10/13/google-gemini-4-argon-api-release/📌 【Google DeepMind】Gemini 4 Argon GA,單次輸出衝到 100 萬 token
TL;DR:Google 旗艦模型 Argon 正式開放 API,一次回應最高可寫 1M tokens,主打長流程 agent 任務。
多數 LLM 的單次輸出被鎖在幾萬 tokens 的量級,逼著開發者把長任務拆成一輪又一輪的呼叫。Google DeepMind 剛把這道牆往上推了超過 15 倍——Gemini 4 Argon 現在一次回應最多可以寫到 1M tokens。
🤔 從小範圍測試到全面開放
Gemini 4 Argon 是 Google 打造的 sparse mixture-of-experts 前沿模型,專為長、多步驟的工作設計,目標領域包括程式開發、金融、法律、科學與網路安全。它最早在 9 月 30 日對外公告,僅開放給一小群網路安全防禦人員使用;現在則全面進入付費 Gemini API,面向開發者、企業與付費 Gemini 訂閱戶。對開發者而言,最大的變化就是輸出長度:Argon 支援的最大輸出從早期 Gemini 模型的 64K tokens,一口氣拉高到 1M tokens。
🧩 新模型,不是微調版
根據官方 model card,Argon 是全新訓練的模型,並非既有 Gemini 的 fine-tune。它是大規模 sparse MoE transformer,具備原生多模態支援:輸入可接受文字、圖片、音訊與影片,輸出則為文字。Google 使用 TPU、搭配 JAX 與 ML Pathways 框架訓練,知識截止日期為 2026 年 3 月。官方並未公開總參數量或啟動參數量。
需要注意的是,model card 說明輸入加輸出的合併視窗上限為 1M tokens——也就是說,提示詞越長,留給輸出的空間就越少。針對長時間解碼,Google 在開發者文件與企業文件中也說明了一套新的「continuation protocol」。以導入期定價計算,一次寫滿 1M tokens 的輸出,光是輸出端就要花費 10 美元。
📊 多數指標領先,但評測方法值得留意
Google 將 Argon 與 GPT-6 Astra、Claude Fable 5.1、Claude Opus 5.5 做比較,在其公開的對照表中,Argon 在多數項目上取得最高分,另外在 ERQA 具身推理(embodied reasoning)benchmark 上拿下 87%。不過官方自己也提醒,應該先閱讀其評測方法說明再下結論:多數對手模型的分數是自報(self-reported)或取自公開排行榜,例如 Argon 的 DeepSWE 分數是用 mini-swe-agent harness 自行計算,而對手的 DeepSWE 數字則來自各家的 system card 與排行榜,評測條件並不完全一致。Google 也承認,在其自家對照表中,Argon 在 5 個項目上落後對手。
⚠️ 已知限制與安全框架
Model card 列出的限制包括幻覺(hallucination)與偶發的回應變慢或逾時。在 Google 的 Frontier Safety Framework 下,官方表示無法排除 Argon 已觸及生物與網路安全相關的「critical capability」等級,但已部署其評估後認為能降低該風險的防護措施。
💡 定價:比 GPT-6 Astra 便宜,打平 Claude Opus 5.5
導入期定價為每百萬 tokens 輸入 2 美元、輸出 10 美元,快取輸入享 95% 折扣。2026 年 11 月 30 日後,價格將調整為每百萬 tokens 輸入 4 美元、輸出 20 美元。這個價位與 Claude Opus 5.5 相同,低於 GPT-6 Astra 的每百萬 tokens 輸入 10 美元、輸出 50 美元。
🎯 實務啟示
如果你的應用屬於「一次性產出長文件」型態——例如完整的程式碼庫重構、長篇法律或科學文件生成——1M tokens 的單次輸出能大幅減少多輪呼叫與拼接的工程負擔。但在評估 Argon 是否優於現有方案前,務必對照 Google 自己提供的評測方法說明,因為跨模型比較的分數來源並不一致,直接拿排行榜數字做決策容易失真。
🔗 來源
- 標題:Google Releases Gemini 4 Argon API With 1M Token Output Limit
- 作者/機構:Asif Razzaq, MarkTechPost
- 連結:https://www.marktechpost.com/2026/10/13/google-gemini-4-argon-api-release/
#GoogleDeepMind #Gemini #LLM #MoE #AIAgents #FrontierModel #AIAPI #LongContext #AISafety #GenerativeAI
原始資料 MarkTechPost · 收集於 2026-10-14
摘要原文
Google DeepMind has made Gemini 4 Argon generally available to developers, enterprises and paid Gemini subscribers. Gemini 4 Argon is a sparse mixture-of-experts frontier model from Google built for long, multi-step work. Its target domains are coding, finance, legal, science and cybersecurity. Google first announced Argon on September 30 for a small group of cyber defenders. It now reaches the paid Gemini API at $2 input and $10 output per million tokens. The key change for developers is output length: Argon can write up to 1M tokens in one response. According to the model card , Argon is a new model, not a fine-tune of an earlier Gemini. It is a large-scale sparse MoE transformer with native multimodal support. It accepts text, images, audio and video, and it outputs text. Google trained it on TPUs using JAX and ML Pathways. The knowledge cutoff is March 2026. Google has not disclosed total or active parameter counts. Google raised the output limit to 1M tokens, up from 64K on earlier Gemini models. The model card describes a combined input plus output window of up to 1M tokens. So a long prompt reduces the room left for output. Google documents a new continuation protocol for long decodes in its developer docs and enterprise docs . At introductory pricing, a full 1M token response costs $10 in output tokens. Google compares Argon with GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5. Argon posts the top score on most rows. Google also reports 87% on the ERQA embodied reasoning benchmark . Read the evaluation methodology before drawing conclusions. Most rival scores are self-reported or pulled from public leaderboards. Argon’s DeepSWE score is self-computed with a mini-swe-agent harness. The rival DeepSWE numbers come from system cards and a leaderboard. Argon trails on 5 rows of Google’s own table: The model card also lists hallucinations and occasional slowness or timeouts. Under Google’s Frontier Safety Framework, Google says it cannot rule out the biological and cybersecurity critical capability levels. It reports deploying safeguards it assesses as lowering that risk. Google shared internal results in its launch post : Introductory pricing is $2 input and $10 output per million tokens. Cached input tokens are 95% off. After November 30, 2026, the price rises to $4 and $20. That matches Claude Opus 5.5 and sits below GPT-6 Astra’s $10 and $50. Access paths, per the developer post : Check out the Model Card , Evaluation Methodology and Developer Guide . 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. The post Google Releases Gemini 4 Argon API With 1M Token Output Limit appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成