Grounded citations with bounding boxes make extracted values auditable, which matters for compliance and back-office automation where a wrong field is costly. Teams processing permits, invoices or scanned forms can benchmark Extract v2.5 against a frontier-model pipeline and likely cut spend by 30% to 4x at higher accuracy. It is a direct challenge to using raw frontier models for document workflows.
OpenAI 新发的 GPT-6.1 Sol (Max) 把价格砍到 GPT-6 Astra 的两成，性能只差 1 分左右，做智能体选模型的可以看看这份榜单。
This release demonstrates that specialized AI agents can now surpass general-purpose models on specific tasks while being significantly more cost-effective. Developers and enterprises working with document processing should evaluate these agents against their current solutions, especially given the substantial accuracy gains on complex documents. The pricing advantage makes it particularly attractive for organizations processing large volumes of documents.
OpenAI 新版 GPT-6.1 Sol 性价比飙升，性能接近 Astra 但只需五分之一价格，各领域排名全面上升。
OpenAI新出的GPT-6.1 Sol，性能接近旗舰Astra但只要1/5价格，适合预算有限但需要高性能的开发者。
Sonnet 5.5 速度比 Opus 快 30%，成绩还追平自家旗舰，价格没涨，干活的性价比党可以直接看看。
Sonnet 5.5 编程测试居然比 Opus 5.5 还高 4 分，价格只要 $2/$10，但开了 max 档费用会飙升，用的时候注意。
OpenAI 把停了半个月的 200 美元 Pro 档重新放开了，额度算下来翻倍，还去掉了 5 小时限制，重度用户可以看看要不要上车。
有人拿非生成式模型 Jev 1.13 和 GPT-6 Sol 在四个医疗基准上跑了对比，便宜快十倍但复杂病例上差 20 多个百分点，概率校准数据挺有意思。
用 Sonnet 5.5 前先看这篇：max 深度下 token 消耗几乎和 Opus 5.5 一样，选档位前算算账。
OpenAI 把 GPT-6 Sol 短暂开放到 Arena 的 Direct Mode，想去试试新模型手感的可以抓紧这 24 小时。
Safety refusals, not coding skill, now decide leaderboard rankings on defensive security work, a factor most agent benchmarks ignore. Security teams choosing models for vulnerability triage should check refusal rates alongside raw scores, since a top coder may decline a third of real tasks. The ranking also spans Chinese and US labs, making it a useful cross-vendor comparison point.
一天里这么多动态：GPT-6修了图像bug、Claude Sonnet 5.5标签被扒出来、Nano Banana 2.1现身Flow，编译排结果都在这。
Anthropic 下一款 Sonnet 5.5 被内测网友扒出来了，实测编码干过 GPT-6 Sol，价格只要 $2/百万 token，文章还教你怎么检查自己有没有被灰度到。
如果你用 GPT-6 Sol 或 Luna 做过视觉任务觉得变差了，现在可以去 API 或 Codex 里重新试试，OpenAI 已把 bug 修好。
两家同天发新模型：GPT-6 便宜一半，Opus 5.5 便宜四成还快三成，还有 Claude 应用商店，价格党可以直接对比一下。
OpenAI 的 Ultrafast 模式能跑出每秒 750 个 Token，比标准模式快 14 倍，现在文档里出现了扩大开放的迹象，可能 DevDay 就官宣。
Routing model selection and reasoning effort to a lightweight System One model shifts cost and latency decisions from static configuration to per-request inference, a pattern that compounds savings on agentic workloads with many calls. Builders writing custom harnesses, especially with the Pi SDK, can plug the router into OpenRouter and reuse the shared implementation to benchmark it against their own fixed-model baseline before committing.
OpenAI 新模型在 Agent Arena 用一半价格逼近 Claude Fable 5 的表现，做智能体应用的成本压力小多了
