模型发布/更新
-
Anthropic releases Claude Opus 5.5, 40% cheaper than Opus 5
Anthropic released Claude Opus 5.5, the first model in the Claude 5.5 series. The company says it performs comparably…
-
OpenAI releases GPT-6 Sol and GPT-6 Luna, API prices 50% lower than GPT-5.6
OpenAI released GPT-6 Sol and GPT-6 Luna, applying the training methods of GPT-6 Astra to faster and cheaper models.…
-
Claude Opus 5.5 发布:较 Opus 5 降价提速,系统卡披露安全演习中约半数运行或有危害行为
Anthropic 发布 Claude Opus 5.5,称达到 Fable 5.1 级性能,相较 Opus 5 输入/输出价格降至 $4 和 $20 每 1M tokens,缓存读取降 60% 至 $0.20,输出提速超 30%,Fas…
-
GPT-6 Sol 与 GPT-6 Luna 在 ChatGPT Work 和 Codex 中推送
OpenAI 的 ChatGPT 官方账号宣布 GPT-6 Sol 和 GPT-6 Luna 两款模型即日起推送,面向 Plus、Pro、Business、Enterprise 和 Edu 用户,可在 ChatGPT Work 和 Cod…
-
Claude Opus 5.5 上线 OpenRouter,智能体编码等能力领先 Opus 5
Anthropic 的 Claude Opus 5.5 上线 OpenRouter,是 Claude 5.5 系列首个模型,在智能体编码、知识工作和计算机使用上领先 Opus 5 和 Fable 5.1。提供 1M 上下文窗口,定价为每百…
-
Qwen-Image-2.1 开源发布,登顶 Arena 图像编辑榜开源第一
通义千问发布 Qwen-Image-2.1,开放权重,在 Arena Image Edit Arena 以 1367 分位列开源第一、总榜第 16,距第 15 名 GPT-Image-1.5-high-fidelity 仅 3 分,同时登…
-
Boris Cherny 实测 Claude Opus 5.5:比 Fable 5.1 快且便宜 51%
Anthropic 的 Boris Cherny 称 Claude Opus 5.5 近几周是他的日常主力模型。他让 Opus 5.5 和 Fable 5.1 各自把 HAProxy 从 C 移植到 Rust,两者都通过了几乎所有测试,但…
-
Anthropic 发布 Claude Opus 5.5,Claude 5.5 系列首个模型
Anthropic 发布 Claude Opus 5.5,是 Claude 5.5 系列的首个模型。官方称其在多数任务上达到 Claude Fable 5.1 的水平,运行成本比 Opus 5 低 40%。
产品发布/更新
-
transformers now loads GGUF quantized models directly, with local inference performance close to llama.cpp
Hugging Face announced that transformers can now run GGUF quantized models directly, loading GGUF checkpoints from the…
-
OpenRouter launches Batch API with half-price batch inference
OpenRouter released the Batch API, in which asynchronous batch requests are completed by providers within a 24-hour…
-
Kimi releases browser extension, renamed from Kimi WebBridge
Kimi released the new Kimi Browser Extension, formerly Kimi WebBridge. Users can chat with Kimi in the browser sidebar…
-
Claude Code v2.1.280 released, adds Claude Opus 5.5 as default Opus model
Claude Code released v2.1.280, adding Claude Opus 5.5 (claude-opus-5-5) as the default Opus model with 1M context…
-
LiteParse September update: PDFium 20-25% faster, new visual grounding and is-complex routing API
LlamaIndex released LiteParse 2.14.6, cutting text extraction time by 20-25% to an average of 2.76ms per page and…
-
Apple's new Mac mini (M6/M5 Pro) and Mac Studio (M5 Max/M5 Ultra) go on sale today
Apple announced that the new Mac mini and Mac Studio go on sale on September 22. The Mac mini features M6 and M5 Pro…
-
OpenAI 为 GPT-6 推出改进的提示词缓存系统与诊断工具
OpenAI 为 GPT-6 系列推出改进的提示词缓存系统,默认提高缓存命中率,对 30 分钟窗口内复用的合格共享前缀提供最高 90% 的缓存输入 token 折扣。
行业动态
-
Pentagon internal review: overreliance on Palantir Maven AI system led to mistaken strike on Iranian school, killing 123 children
According to Bloomberg, citing unnamed officials involved in an unreleased Pentagon internal review, a mistaken U.S.…
-
Claude Opus 5.5 added to Arena's Agent Arena and Battle Mode
Arena announced that Anthropic's Claude Opus 5.5 has entered Agent Arena, where user votes drive the leaderboard.…
技巧与观点
-
Claude Opus 5.5 tops the Artificial Analysis Intelligence Index
Artificial Analysis measured Claude Opus 5.5 at a score of 58 on the Intelligence Index, its highest recorded score,…
-
Artificial Analysis evaluates GPT-6 Sol and Luna: prices halved but performance varies
Artificial Analysis evaluated OpenAI's GPT-6 Sol and Luna, whose prices are roughly half those of GPT-5.6, with Sol…
-
Kazike tests Grok 4.7 against Xiaomi MiMo V2.6, calling the latter the answer to the impossible triangle
The author tested Grok 4.7 and Xiaomi MiMo V2.6, both released in the early hours of the same day, and found Grok 4.7…
-
Artificial Analysis evaluates Step 5 Preview: Intelligence Index score of 44, cost about 1/2.8 of peer models
Artificial Analysis evaluated StepFun's Step 5 Preview, which scored 44 on the Artificial Analysis Intelligence Index,…
-
OpenRouter tests Jev 1.13 vs Claude Opus 5 on Banking77 classification accuracy, latency and cost
OpenRouter compared Jev 1.13 and Claude Opus 5 on intent classification using 3,080 customer service samples from the…
-
OpenRouter explains how NVIDIA Nemotron 3.5 Lightning handles high-frequency agent execution calls
OpenRouter published an analysis of NVIDIA's Nemotron 3.5 Lightning, a 30B total-parameter, roughly 3B…