大模型
x.com ·
@TheRundownAI
The Rundown AI 转述 OpenAI 发布 GPT-6.1 Sol 及多项产品
The Rundown AI 发帖称 OpenAI 发布 GPT-6.1 Sol，宣称以约五分之一价格接近 Astra 智能水平：DeepSWE 编码与 GPT-6 Astra 持平、AutomationBench 办公流程领先 Opus 5.5 2.2 分、OSWorld 2.0 计算机操作落后 Astra 2.1 分；定价为输入 2 美元/百万 token、输出 10 美元/百万 token。同时提及 Pro 500 套餐、Codex 的 Ultrafast 加速以及 Luna 驱动的 Decisions API。上述数据均来自该帖转述，未经官方来源核实。
AI推荐理由 该帖转述 OpenAI 发布 GPT-6.1 Sol 及多项产品，给出 DeepSWE、AutomationBench、OSWorld 2.0 的对比数字和定价，信息密度较高；但来源为第三方账号转述，未提供官方链接或可验证证据，且模型名与基准无法独立核实，故质量分受限。对关注模型能力与成本趋势的读者有参考价值。
62
浏览 3,138
点赞 9
评论 3
转发 2
正文
GPT-6.1 Sol is here, with OpenAI promising "near-Astra intelligence for a fifth of the price."
Some numbers:
- Real-world coding (DeepSWE): ties GPT-6 Astra at roughly 1/5 the cost
- Office workflows (AutomationBench): beats Opus 5.5 by 2.2 points at 1/3 the cost
- Computer use (OSWorld 2.0): 2.1 points shy of Astra at around 1/7 the cost
Pricing: $2 per million tokens in, $10 out.
Also announced:
- Pro 500, a new plan with 25x the usage of Plus, and Ultrafast, up to 8x faster speeds in Codex
- Decisions API, near-instant multiple-choice calls from Luna for sorting content and routing agents
The Rundown AI@TheRundownAI
The Dev Day headliner: Dots.
OpenAI's answer to Muse / Grok Bot, AI agents that get their own computer and keep working independently.
Runs on GPT-6 Astra, connects to 4,000+ apps, and reachable in ChatGPT, Slack, and Teams.
[Your browser does not support the video tag.](https://video.twimg.com/amplify_video/2104979273312583680/vid/avc1/1920x1080/vbvkgmNi7bZAstnb.mp4?tag=29)
[🔗 View on Twitter](https://x.com/OpenAI/status/2104980481876070819/video/1)
[🔗 View Quoted Tweet](https://x.com/TheRundownAI/status/2104981371961974797)