X

We built a patient intake agent with @SpaceXAI that runs end to end on Grok voice models. Grok STT → Grok 4.3 → Grok TTS, cascaded through LiveKit Inference. Three model strings in one AgentSession. No separate API key, no separate billing, ZDR on every hop. Talk to it: https://xai.livekit.space

牛来大模型刚被彭博社破案了！ 在 OpenRouter 上悄悄屠榜、调用量超过 DeepSeek 一倍的神秘模型 Ox Alpha， 背后竟然是智谱，更绝的是官方确认今晚直接开源权重， 狂刷了数十万亿 token、在真实 Coding 里跑出 80% 峰值，智谱这次是真的要杀疯了， 前阵子大家都在猜这个匿名模型到底是哪家大厂的马甲， 因为它在排位赛上狂刷了数十万亿 token， 开发者用脚投票把它送上了第一， 它真正狠的地方，是把 Flash 级别的速度和成本，做出了中上游前沿模型的实战能力: 1M 超长上下文、原生支持图片和视频多模态， 而且是推理优先的架构，写代码时会先做规划再调工具 社区独立跑测试，在 10 个高难度真实 Coding 任务里拿下了 80% 的过关率， 全量 DeepSWE 跑出 64.6%， 直接咬住了 Claude Opus 4.8 和 Gemini 3.7 Flash 的身位 它不是那种打虚空跑分的绝对 SOTA， 但一个成本极低、速度极快、能看懂视频还能写复杂代码的 Flash 模型， 今晚一旦把权重放出来，无论是本地部署还是第三方 API，生态杀伤力完全不可同日而语 刚发完开源顶流 GLM-5.3，转头就把这个高频 Agent 生产力核弹开源出来， 大模型的竞争，终于从比拼谁的参数更大，彻底转向谁能让开发者用最低成本把活干完了啊

Finally, Z .ai revealed that Ox Alpha was actually GLM-5.3-Flash. So that means over the last few days all those 100 tn tokens/day of stealth traffic capacity was running on Chinese AI chips, with tens of thousands of domestic accelerators behind the service. not an NVIDIA GPU cluster. 5.3-Flash beats GLM-5.2 at one-tenth the price with only 18B active parameters. GLM-5.3-Flash. has 320B params in total, but only 18B are active during inference. It also uses 45 layers instead of GLM-4.5's 92, cutting the amount of work required for each token. The benchmark jumps are large too: against GLM-5.2, DeepSWE rises from 46.2 to 63.4 and AutomationBench from 26.2 to 48.8. There is an architectural change as well, that cuts attention compute 3x and per-layer KV cache 4.4x versus GLM-5.3. The novelty is mainly in the combination: GLM-5.3-Flash uses linear attention for cheap state tracking, then sparse attention with a lightweight indexer to retrieve only the distant context worth revisiting, instead of repeatedly attending across the full 1M-token window. They also introduces IndexPool, which compresses four indexer key vectors into one, and says the combined design

Brain is our self-improving memory system for Perplexity Computer. It compiles sessions, files, and sources into a structured knowledge wiki. New evals build on our initial results, improving correctness by 9.3 points, currentness by 8.0, and recall by 8.9 with 15% fewer tokens. https://t.co/mDMVWt2xzS

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight! The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens. 125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency. What's new: 🥳 - Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4. - Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks. - Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI). - 262K native context, extensible to 1M with YaRN. We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀 We can't wait to see what you build with Qwen3.8-Flash!👀👇 - Blog: https://t.co/M5hYypFLgJ - Technical Report: https://t.co/IF0gObIkQO - Hugging Face: https://t.co/6ow8QVAABt - ModelScope: https://t.co/tDOn2jNuFG

OxAlpha is a new iteration of GLM, from China’s Z .ai And it will change how you run long running agent fast. --- bloomberg .com/news/articles/2026-08-26/china-s-z-ai-made-ox-alpha-stealth-model-that-rivals-deepseek https://t.co/ZOTlnBgOsy

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight! The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens. 125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency. What's new: 🥳 - Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4. - Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks. - Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI). - 262K native context, extensible to 1M with YaRN. We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀 We can't wait to see what you build with Qwen3.8-Flash!👀👇 - Blog: https://t.co/M5hYypFLgJ - Technical Report: https://t.co/IF0gObIkQO - Hugging Face: https://t.co/6ow8QVAABt - ModelScope: https://t.co/tDOn2jNuFG