Hy4 achieves parity with Grok 4.7 at one-tenth the cost, with a Cost per Intelligence Index Task of US$0.78 versus far pricier Western peers. The main caveat is an 89% hallucination rate on AA-Omniscience v1.1 against Kimi K3.1 Pro's 37%, so agentic coding is the safer use case than knowledge retrieval. Teams benchmarking China-based models against Claude, Grok or Qwen options should weigh that trade-off directly.
百度放了个开源 MoE 模型，280B 总参但每个 token 只激活 14B，跑得省，官方说同量级里能打 SOTA，完整的 5.5 Flash 还在路上。
腾讯混元 Hy4 要上 Chutes 了，用你现在那把 API key 换个模型 ID 就能调用，用 Chutes 的朋友留意下。
腾讯把 Hy4 接进了 OpenClaw，1M 上下文加视觉，做长任务智能体的可以试试，走 OpenRouter 就能用上。
OpenAI 新出的 GPT Image 2.6 两个版本，出图最快 5 秒，文生图榜还包揽前五，编辑一致性也升级了。
OpenAI 发了 GPT-Image-2.6，生成快了 50%，Sunburst 版本编辑图片时细节更稳，走 API 的可以试试。
OpenAI 给开发者上新了两个图像模型：Flare 6 秒出图，Sunburst 质量更高还支持 SVG 导出，做图像应用可以试试。
Perplexity 的 decider 模型三周内流量涨到 Jev 的十分之一，OpenRouter 数据能直接查，可以看看这匹黑马。
65B 的小模型居然在编程智能体任务上打过了 800B 级别的大家伙，美国开源阵营又有新选择。
月之暗面把 K3.1 Pro 权重放出来了，Agent Arena 上开源第一，Shell 命令成功率和听指挥能力都涨了不少，玩开源智能体的可以试试。
五人小团队从零训了个 65B 稠密模型，拿了美国开源在 Terminal-Bench 2.1 的最好成绩，训练细节写得挺实在。
The notable detail is the lineage: a coding-agent company building on Thinking Machines Lab's open Inklingsmall suggests smaller open base models can compete with in-house large models on agentic coding at a fraction of the cost. Teams running coding agents should benchmark SWE-2 Flash against SWE-2 and alternatives like Claude-based agents, since the 5× price cut and 3× speed change the cost equation for high-volume agent workloads.
Day-0 availability on Databricks signals Moonshot pushing Kimi into enterprise data platforms rather than consumer chat apps, putting it next to 70+ alternatives where procurement teams actually evaluate models. Unity Gateway means companies can A/B Kimi against rivals without code changes — useful for cost benchmarking. Enterprises with Lakehouse data estates, including those in Singapore and ASEAN running Databricks, can try it immediately.
月之暗面把 Kimi K3.1 Pro 权重放到 HuggingFace 了，3T 参数的多模态智能体模型，能自己下载部署。
A 65B dense model beating 500B-plus open-weight rivals on agentic coding would reshape assumptions about how much capacity agents actually need. Teams choosing self-hosted coding models should benchmark Ripple against Nemotron 3 Ultra-550B and the other named releases before committing to larger deployments. No benchmark names or independent evaluations are given yet, so treat the state-of-the-art claim as the lab's own.
A 65B dense model matching or beating a 501B sparse model on Terminal-Bench 2.1 would shift assumptions about how much parameter count matters for agentic coding. The Apache 2.0 licence and local runnability mean teams can self-host a competitive coding agent without API dependence. Benchmark figures here are company-reported, so independent verification should be the first stop before adopting it.
webAI 开源了 88 亿参数的 TwIL-8B，编码、指令遵循、数学都能打到前沿模型九成左右的成绩，还小到自己的电脑能跑，权重直接放 Hugging Face 上了。
An open-weight 30B MoE with only ~4B active parameters makes speech-to-text with speaker diarisation and translation runnable on modest hardware, undercutting the usual cost of per-language ASR pipelines. Builders working on voice agents or meeting transcription can self-host one checkpoint instead of chaining multiple services. With 40-language coverage, it is a practical candidate for multilingual Southeast Asian speech workloads.
月之暗面把 2.8T 参数的 Kimi K3.1 Pro 权重直接开源了，4-bit 发布省存储，编程基准分数贴近 Claude Opus 5.5，纯知识题却差 GPT-6 一截，许可证对云托管商还有限制。
LMArena 的 Agent Arena 榜单上新了，能看到各智能体模型在不同任务间的取舍对比，选模型前可以先去比一比。
For hour-long agent workloads, the cheapest model is not automatically the right one: Claude Opus 5.5 scores higher but costs twice as much per hour as Gemini 4 Carbon. Teams running long-horizon agents can use these per-hour cost and time figures to match model choice to budget and quality needs. The 55-minute spread among frontier models also suggests task duration, not just accuracy, should enter model selection criteria.
Agent Arena 拆了成本维度看模型：贵任务上 OpenAI 和 Anthropic 不再通吃，Gemini 4 Argon 和小米 MiMo 顶上，选型前值得对号入座。
Cost differences widen precisely on the longest sessions, so the cheapest model for simple tasks is not the cheapest for complex agentic work. Teams running long-horizon agent tasks such as app development can cut bills by re-benchmarking at high task complexity rather than averaging across workloads. The finding also opens the door for Gemini 4 Argon and Xiaomi's MiMo-V2.6 Flash and Pro to challenge OpenAI and Anthropic where budgets are tightest.
Agent Arena 拿 5 家实验室的旗舰模型跑了长任务对比，Claude Opus 5.5 能连续干近 2 小时，Kimi 最快但也最短，差距挺直观。
The result argues that at-scale AI code generation can meet formal verification standards, not just quick prototypes. Mathematics researchers and formal-methods teams gain a machine-checked base they can build on, and it signals AI agents moving from toy proofs to large structured libraries. It also compares favourably with prior partial formalization efforts in scope and coherence.
The zero-data-retention and US-hosting terms matter for teams that couldn't previously route sensitive workloads to a Chinese open-weight model. Developers can now self-host or run K3.1 Pro through Ollama Cloud as an alternative to renting closed APIs from OpenAI or Anthropic. Weight availability on Hugging Face also means inference providers beyond Ollama can deploy it.
Perceptron AI 实测发现标注机器人视频用专用小模型就够，比顶级大模型省 8 倍钱，质量几乎不打折，做机器人的可以看看。
Moonshot AI 出了 Kimi K3.1 Pro，Together AI 上就能直接调用， podcast 里有他们自己人讲细节，想用的话可以先听听。
Moonshot 开放了 Kimi-K3.1-Pro，2.8T 参数只激活 104B，1M 长上下文，BrowseComp 拿了 92.1，权重才 1.56TB，跟 Claude Opus 5.5 掰手腕。
The plug-in design means no retraining of the base model, so teams running open-source models could adopt ZHA by training or downloading a small index module. It matters for anyone serving long-context workloads, where KV-cache reads dominate cost and latency. Compared with MSA, the reported 12.7-point retrieval gain is the concrete benchmark to watch when evaluating sparse attention options.
一个用5300小时视频就练出来的视频识别模型，22个基准平均78.76%，在医疗和遥感任务上还反超了Meta烧几百倍算力的VJEPA，权重已开源。
OpenAI 自己发的 LifeVal 榜单，GPT-6 Astra 比 GPT-5.2 涨了 31 个点，不过任务集没公开，看看就好
腾讯的 Hy4 在 APEX-Agents 上跑赢 DeepSeek V4.1，定价也放出来了，还说要开源权重，开源圈可以盯一下。
字节靠 Seedance 把腾讯、B站变成 API 客户，内容还得在抖音、TikTok 分发，版权又和迪士尼谈妥，商业模式挺值得琢磨。
Only the latest 2000 items are listed; for older ones see the daily archive
