GPT-6.1 Sol評測：價格僅Astra五分之一，效能打幾折？ | GPT-6.1 Sol Review: Astra-Level Power, 1/5 the Price

10月 02, 2026

By Kit 小克 | AI Tool Observer | 2026-10-03

🇹🇼 GPT-6.1 Sol評測：價格僅Astra五分之一，效能打幾折？

OpenAI 在 9 月 29 日的 DevDay 上推出 GPT-6.1 Sol，主打用五分之一的價格打出接近旗艦 GPT-6 Astra 的表現。背景是 Astra 一直沒有全面上線——官方的說謊問題沒解決乾脆沒放出完整版，於是先把 Sol 補上去當主力。對天天要算 token 成本的開發者來說，這是目前最值得關注的 AI 模型新聞。

GPT-6.1 Sol 跑分跟Astra差多少？

官方公布的數字相當具體：在模擬真實專案的 DeepSWE v1.1 測試中，GPT-6.1 Sol 幾乎追平 Astra，成本卻只要五分之一，比上一代 GPT-6 Sol 高出 6.4 個百分點。電腦操作測試 OSWorld 2.0 上，Sol 比前代多拿 7 分，只落後 Astra 2.1 分，花費卻只要約七分之一。科學推理測試 Terminal-Bench Science 的分數直接翻倍，單次任務平均成本 5.47 美元，對比 Opus 5.5 的 23.21 美元與 Astra 的 23.80 美元，差距相當明顯。

定價細節：便宜在哪裡？

API 報價是每百萬輸入 token 2 美元、輸出 10 美元，快取輸入只要 0.10 美元（比標準輸入便宜 95%），快取寫入 2.5 美元——整體是 Astra 標準價 10/50 美元的五分之一。

目前開放給 Plus、Pro、Business、Enterprise、Edu 用戶，在 ChatGPT Work 與 Codex 內使用
一般 ChatGPT 聊天介面還沒有，只能透過 API 或 Codex 接觸
新增 Ultrafast 速度模式，輸出速度約每秒 300 token
適合長時間跑的 agent 任務、大量重複呼叫、需要控制帳單的團隊

GPT-6.1 Sol 值得換嗎？

如果你原本的工作流是中長程式碼修改、電腦操作自動化，GPT-6.1 Sol 用五分之一甚至七分之一的成本拿到九成以上的表現，對大部分團隊是划算的升級。真正需要頂級推理、或任務複雜到逼近 Astra 上限的場景，才需要堅持用更貴的模型。多數日常開發工作，便宜又夠用的 Sol 已經是更實際的選擇。

好不好用，試了才知道。

🇺🇸 GPT-6.1 Sol Review: Astra-Level Power, 1/5 the Price

OpenAI launched GPT-6.1 Sol at DevDay on September 29, pitching it as near-flagship performance at one-fifth the price of GPT-6 Astra. The backstory matters: Astra never got a full public release — OpenAI reportedly couldn't fully fix its lying problem, so Sol became the de facto flagship instead. For any developer tracking token costs, this is the AI model story worth paying attention to right now.

How close does GPT-6.1 Sol get to Astra?

The published numbers are specific. On DeepSWE v1.1, which simulates real-world software engineering work, GPT-6.1 Sol matches Astra's score at roughly a fifth of the cost, beating the previous GPT-6 Sol by 6.4 points. On OSWorld 2.0, a computer-use benchmark, Sol gains 7 points over its predecessor and lands just 2.1 points behind Astra — at about a seventh of the cost. On Terminal-Bench Science, Sol more than doubles its predecessor's score; an average science task costs $5.47, versus $23.21 for Opus 5.5 and $23.80 for Astra.

Pricing: where the savings come from

The API pricing is $2 per million input tokens, $10 per million output tokens, cached input at just $0.10 per million (95% off standard input), and $2.50 per million for cache writes — altogether a fifth of Astra's $10/$50 standard rate.

Available now to Plus, Pro, Business, Enterprise, and Edu users inside ChatGPT Work and Codex
Not yet in the main ChatGPT chat interface — API or Codex access only
Ships with a new Ultrafast mode running at roughly 300 tokens/second
Best fit for long-running agent tasks, high-volume calls, and teams watching their bill

Is GPT-6.1 Sol worth switching to?

If your workflow involves long-running code edits or computer-use automation, GPT-6.1 Sol delivers 90%+ of top-tier performance at a fifth to a seventh of the cost — a straightforward upgrade for most teams. Only workloads that genuinely need peak reasoning, pushing right up against Astra's ceiling, still justify the pricier model. For everyday development work, the cheaper, good-enough Sol is now the more practical pick.

好不好用，試了才知道。

Sources / 資料來源

延伸閱讀 / Related Articles

AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

搜尋此網誌

AI小貼士