Chinese contentGemini 4 临近正式发布，GLM、DeepSeek、Kimi 一段时间没有动静
Gemini 4 快来了，而且 Anthropic 博客都还在引用 GLM-5.3 的网络能力，国产模型这条线索值得关注。
Latest
Every published item, in order · 131 items
Oct 4
Oct 3
GLM-5.3 matches Claude in vulnerability testing
The performance gap between open-source and closed models is narrowing faster than expected, with GLM-5.3 demonstrating competitive vulnerability exploitation capabilities. Developers and security researchers should evaluate this model for ethical hacking and security testing applications, especially given its open weights nature that allows for customization and inspection.
Sep 30
SPLASH system switches attention layouts
SPLASH eliminates the need to restart workers when changing attention layouts, enabling dynamic adaptation to varying workloads. Developers can implement this to optimize LLM serving efficiency across different hardware configurations and model sizes, with particular benefits for handling mixed workloads that transition between short and long requests.
Chinese contentAnthropic测评GLM-5.3：开源模型达前沿水平
Anthropic给中国智谱开源模型GLM-5.3做了测评，说它能力已达前沿水平且几乎不设防，这算提醒还是变相宣传？
Sep 29
Sep 25
TileRT and AMD push GLM-5.3 to 469 tok/s decode on MI355X
A disaggregated prefill-decode split is becoming the practical recipe for low-latency serving, and this run shows AMD hardware competing at interactive agent speeds rather than just batch throughput. Developers benchmarking inference stacks should note that the V1 connector lets third-party decode engines plug into vLLM without forking the stack. The result also gives GLM-5.3 a concrete serving data point against US-frontier-model deployments on NVIDIA GPUs.
Chinese contentArtificial Analysis 推出 Terminal-Bench-Science 0.1 智能体科研基准榜单
Stanford 联合 Terminal-Bench 团队搞了个科研智能体榜单，70 个真实科研任务里 GPT-6 Astra 拿 63% 排第一，开源模型才 10%，差距很直观。
GLM-5.3 Edges Space Bunny Alpha in Physics Simulation Test
Head-to-head physics benchmarks remain rare, and this one suggests GLM-5.3 holds an edge over an unnamed stealth competitor on collision-accurate simulation. Builders testing video or world models can replicate the 5-scene setup to compare candidates on dynamics rather than visual polish. The stealth model's OpenRouter availability makes it easy to verify the result firsthand.
Sep 23
Step 5 Preview tested on two real coding tasks
Benchmark scores can hide whether a model survives long, multi-step builds, so the value here is watching Step 5 Preview handle mechanical synchronisation and real tooling rather than canned puzzles. Developers choosing between Step 5, GLM-5.3, Kimi K3 and Claude Opus 5 get a practical cost-versus-reliability comparison for agentic coding workflows. The stated price gap makes it a concrete candidate for teams in Southeast Asia and beyond cutting per-task inference spend.
Sep 22
Chinese contentXiaomi 发布 MiMo-V2.6，将开源 RL 训练全程
Xiaomi 把 RL 训练全开源，Pro 版跑赢 Qwen3.8 Max，价格比涨价后的 DeepSeek 更低。
Open model releases since July
Chinese models like Qwen3.8-Max and DeepSeek V4-Pro now rival top closed-source options, affecting AI development costs globally.
Chinese content小米开源全模态模型 MiMo-V2.6：AA 指数 46 分超 Kimi K3
小米把 MiMo-V2.6 权重全开源了，Pro 在 AA 指数拿 46 分压过 Kimi K3，价格只有海外模型的 1/20。
IT之家Source
Sep 20
Sep 19
Chinese content智谱 GLM-5.3-FlashX 模型上线，更快、更流畅
智谱新出的 GLM-5.3-FlashX 模型，速度提升明显，适合需要高效处理大量文本的场景。
IT之家Source
Sep 18
Chinese content智谱发布GLM-5.3-FlashX，推理速度达200 tokens/s
智谱AI新出的GLM-5.3-FlashX模型，在国产加速器上跑得特别快，每秒能处理200个token，比之前版本效率高不少。
Zhipu GLM-5.3-FlashX model released
For users needing a fast, high-performance AI model, the GLM-5.3-FlashX offers a competitive alternative to other large language models.
IT之家Source
Sep 17
Zai.org optimizes GLM-5.3-Flash for production
This matters because it demonstrates a practical approach to optimizing large language models for real-world production environments, which is crucial for companies looking to deploy such models efficiently.
Zhipu GLM reveals first RSI practice for domestic large models
This is the first time a domestic Chinese large model company has publicly used AI self-improvement for production environments, showing AI's ability to participate in the entire engineering loop of a reasoning system.
Zhipu optimizes GLM-5.3 inference with its own AI system
For users of Chinese AI models, this shows how a model can autonomously optimize its own performance, potentially leading to faster and more efficient updates in the future.
Xiaomi trains MiMo-2.6 model
For users looking for affordable AI models, mimo-v2.5's strong performance on OpenRouter and mimo-v2.6's improved results show Xiaomi's models remain competitive in price and capability compared to alternatives like GLM-5.3-Flash.
Sep 16
Chinese contentImage-to-WebDev Arena更新：四款新模型发布
OpenAI发布了GPT-6 Astra (Max)，在Image-to-WebDev任务上表现最好，比GPT-5.6 Sol好很多。
Sep 12
Chinese contentDeepSeek V4.1 Flash 在 OpenRouter 上 24 小时处理 1T token
DeepSeek 新模型 V4.1 Flash 性价比很高，在 OpenRouter 上 24 小时处理 1T token，比 GLM-5.3 Flash 便宜 5 倍。
Sep 11
Chinese contentGLM-5.3 可在 Fireworks 训练 API 上训练
朋友，Fireworks 新增了 GLM-5.3 的训练支持，这个模型适合做复杂编码和长时程代理，你可以在他们的专用平台上试试。
Chinese contentDeepSeek 发布 V4.1 Flash 模型，性能对标 GLM-5.3，KV 缓存技术大幅降低内存占用
DeepSeek 新出的 V4.1 Flash 模型，性能不错，和 GLM-5.3 差不多，关键是内存消耗特别低，适合做 agentic 任务，价格也便宜很多。
Sep 8
Chinese contentOx Alpha模型实为Zai的GLM-5.3-Flash
Zai的GLM-5.3-Flash模型就是之前爆火的Ox Alpha，320B参数但成本仅0.09美元，还用国产硬件跑通了。
Sep 7
Sep 4
GLM-5.3 and DeepSeek-V4 models
These models offer competitive alternatives to Western AI systems with strong performance in Chinese language processing.
Sep 2
Chinese contentGLM-5.3-Flash首秀进代码榜前十，千问3.8-Max登顶前端榜
国产模型在细分领域表现亮眼，GLM-5.3-Flash代码能力超越多数海外模型，千问3.8-Max直接登顶前端榜。
Sep 1
Chinese contentGLM-5.3-Flash 加入 Agent Arena
智谱AI的GLM-5.3-Flash在Agent Arena表现亮眼，成本仅0.12美元/任务，超越自家前代模型。
GLM-5.3-Flash ranks #19 overall
GLM-5.3-Flash offers strong performance among open models, making it a competitive alternative for developers in ASEAN seeking accessible AI solutions.
Aug 31
Aug 29
GLM-5.3-Flash model released
GLM-5.3-Flash offers developers in ASEAN access to Zhipu's latest model through Fireworks' platform.
Zhipu releases GLM-5.3 model weights
GLM-5.3 offers advanced AI capabilities for coding and cybersecurity at lower cost than closed-source alternatives, with commercial use permitted under specific terms.