OpenAI Releases GPT-6 Sol and Luna: 50% Cheaper API Pricing and Benchmarks
https://www.marktechpost.com/2026/09/22/openai-releases-gpt-6-sol-and-luna-50-cheaper-api-pricing-and-benchmarks/📌 【OpenAI】GPT-6 Sol、Luna 上線：價格砍半，部分項目打贏 Opus 5
TL;DR：GPT-6 Sol 與 Luna API 價格較 GPT-5.6 促銷價降 50%，多項 benchmark 顯示成本效益大幅優於 Claude 對手。
一個模型分數贏過對手，但成本只要對方的十分之一，這樣的組合才是企業真正在意的東西。OpenAI 這次發布的 GPT-6 Sol 與 Luna，主打的不是「更聰明」，而是「用更少的錢做到接近，甚至超越的表現」。
🤔 GPT-6 家族補齊中低階產品線
繼本月稍早發布頂規模型 GPT-6 Astra 之後，OpenAI 再推出 Sol 與 Luna 兩款定位在 Astra 之下的模型。三者訓練方式類似，目標是把 Astra 的進展帶進更快、更平價的版本。目前 GPT-6 家族分為三層：Astra 負責最困難的任務；Sol 鎖定複雜的程式開發與專業工作，成本較低；Luna 則主打快速、高流量的日常任務。兩款模型皆已在 OpenAI API 中上線（gpt-6-sol、gpt-6-luna），僅提供 API 存取，不釋出權重供自行部署。
🧩 更好的快取與推論效率支撐降價
OpenAI 表示，更好的 caching 與推論技術讓這兩款模型能以更低成本服務，因此將 Sol 與 Luna 的 API 定價相較 GPT-5.6 促銷價下砍 50%。值得注意的是 Luna 的輸出價格從每百萬 token $1.20 降至 $0.50，實際降幅約 58%，比官方宣稱的 50% 更多。
📊 跨多項 benchmark 的成本效益比較
| Benchmark | 模型（Effort） | 分數 | 成本備註 |
|---|---|---|---|
| AutomationBench 1.0.6 | GPT-6 Sol（xhigh） | 33.2% | $0.27／task |
| AutomationBench 1.0.6 | Claude Opus 5（max） | 26.9% | 11.1 倍 Sol 成本 |
| AutomationBench 1.0.6 | GPT-6 Astra（low） | 30.3% | 3.9 倍 Sol 成本 |
| Agents’ Last Exam | GPT-6 Sol（max） | 56.4% | 較 Claude Opus 5 最佳成績低 60% 成本 |
| DeepSWE v1.1 | GPT-6 Sol（max） | 68.8% | 較 Claude Fable 5（xhigh）低 1.1 分，成本低約 80% |
| DeepSWE v1.1 | GPT-6 Luna（max） | 66.6% | 與 Opus 5、Fable 5（medium）相當，成本分別低 93%／96% |
| OSWorld 2.0（offline） | GPT-6 Sol（xhigh） | 60.5% | 對比 Opus 5（medium）60.3%，成本低約 80% |
此外，Luna 在 high effort 下較前代提升 5.4 分，成本卻降低 58%；在 FrontierCode 1.1 Main（評估程式碼是否可直接合併）上，Sol 的表現追平 Claude Fable 5.1（xhigh），但成本明顯更低。Luna 在 max effort 下也能以十分之一的成本打贏 GPT-5.6 Sol（medium）於 OSWorld 2.0 的表現。
📊 事實準確度與溝通風格也有進展
在 OpenAI 使用去識別化 ChatGPT 對話（使用者標記錯誤）所做的內部測試中，Sol 犯錯機率約為前代的一半；Luna 在較高 effort 下，準確度可與 GPT-5.6 Sol 打平，成本卻只要約百分之一。OpenAI 也將 Astra 的溝通風格延續到 Sol 與 Luna，回答預期更清楚、稍短、少術語，尤其在程式相關對話中更明顯。
🧩 更聰明的 prompt caching 同步上線
配合這次發布，GPT-6 也預設提供更高快取命中率的 prompt caching 系統，命中快取的輸入最高可折扣 90%，且在 30 分鐘視窗內重複使用的共用前綴（shared prefix）也符合資格。根據報導，GitHub 表示這項改動讓需要重新處理的 prompt token 比例減少超過 50%，有助於 Copilot 更快回應。
🎯 實務啟示
如果你的應用場景是高流量、對延遲敏感的任務（如客服、程式輔助），Luna 的成本效益值得優先評估；若需要更強的推理與 agentic 能力但仍要控制成本，Sol 在多項 benchmark 上展現出比 Claude Opus 5 更划算的表現。導入前建議針對自己的任務跑一輪基準測試，畢竟公開 benchmark 分數未必完全反映實際工作負載。
🔗 來源
- 標題：OpenAI Releases GPT-6 Sol and Luna: 50% Cheaper API Pricing and Benchmarks
- 作者／機構：Sana Hassan／MarkTechPost
- 連結：https://www.marktechpost.com/2026/09/22/openai-releases-gpt-6-sol-and-luna-50-cheaper-api-pricing-and-benchmarks/
#GPT6 #OpenAI #LLMBenchmark #AIÐPricing #ClaudeOpus #AgenticAI #CodingAssistant #APIeconomics #MachineLearning #LLMOps
原始資料 MarkTechPost · 收集於 2026-09-23
摘要原文
OpenAI has released GPT-6 Sol and GPT-6 Luna , 2 new models in its GPT-6 family. They sit below GPT-6 Astra , which launched earlier this month. OpenAI trained both with methods similar to Astra’s. The aim is to bring Astra’s advances to faster, more affordable models. Deployable today? Yes. Both models are live in the OpenAI API as gpt-6-sol and gpt-6-luna . They are API-only models, so there are no weights to self-host. The GPT-6 family now has 3 tiers. Astra is the top model for the hardest work. Sol targets complex coding and professional tasks at lower cost. Luna targets fast, high-volume everyday work. OpenAI team states better caching and inference let it serve these models more cheaply. It is cutting Sol and Luna API prices by 50% against their GPT-5.6 promotional pricing. One detail is worth noting. Luna’s output price falls from $1.20 to $0.50, a cut of about 58%, not 50%. Professional work : On AutomationBench 1.0.6 , Sol at xhigh effort scores 33.2% at $0.27 per task. Claude Opus 5 at max effort scores 26.9% at 11.1x that cost. Low-effort Astra scores 30.3% at 3.9x Sol’s cost. Luna at high effort gains 5.4 points over its predecessor at 58% lower cost per task. On Agents’ Last Exam , Sol at max effort scores 56.4%. That beats Claude Opus 5’s best score at 60% lower cost per task. Coding : On DeepSWE v1.1 , Sol at max effort scores 68.8%. That is 1.1 points behind Claude Fable 5 at xhigh, at about 80% lower cost per task. Luna at max effort scores 66.6%, comparable to Opus 5 and Fable 5 at medium effort. In those comparisons, Luna costs 93% less per task than Opus 5 and 96% less than Fable 5. On FrontierCode 1.1 Main , which grades whether code is ready to merge, Sol matches Claude Fable 5.1 at xhigh at much lower cost. Computer use : On OSWorld 2.0 offline, Sol at xhigh scores 60.5% versus 60.3% for Opus 5 at medium. Sol’s cost per task is about 80% lower. Luna at max beats GPT-5.6 Sol at medium for 1/10 of the cost. Factuality : OpenAI’s internal test uses de-identified ChatGPT conversations where users flagged model errors. Sol makes about half as many mistakes as its predecessor. Luna at higher effort matches GPT-5.6 Sol at about 1/100 of its cost. OpenAI also carried Astra’s communication style over. Expect clearer, slightly shorter answers with less jargon, especially in coding conversations. Agents resend the same instructions, tools and history on every turn. GPT-6 ships an improved prompt caching system with higher cache hit rates by default. Cached input reads get discounts of up to 90%. Eligible shared prefixes reused within a 30-minute window now qualify. The full prompt caching guide covers each pattern. GitHub reports these changes cut the share of prompt tokens needing fresh processing by more than 50%, helping Copilot respond faster. Check out the Technical Blog . 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 OpenAI Releases GPT-6 Sol and Luna: 50% Cheaper API Pricing and Benchmarks appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成