Pikafish 2026-09-06 Github搬运

[font=&quot]今天，我们很高兴宣布发布 Pikafish 2026-09-06 版本。一如既往，您可从 GitHub 发布页面免费下载该版本，并将其作为任意图形界面的即插即用替代品，以获得更强的对局表现和更精准的分析。[font=&quot]无论您能贡献数小时还是数天的 CPU 时间，您的帮助都将推动 Stockfish 的持续开发，同时也巩固了 Pikafish 所依赖的技术基础。请访问 stockfishchess.org/get-involved 了解如何参与贡献。加入他们的 Discord 服务器，与项目开发者及用户社区建立联系！[font=&quot]象棋对弈质量  [font=&quot]在与 Pikafish 2026-01-02 版本的测试中，本次新版本的 Elo 增益高达 24 点，赢得的对局对数超过输掉的两倍。[font=&quot]开局库：胜率 65.85%[font=&quot]时间控制：10+0.1  [font=&quot]总场次/胜/平/负：2766 / 830 / 1421 / 515  [font=&quot]PTNML：7 / 193 / 688 / 468 / 27  [font=&quot]胜率：55.69%  [font=&quot]Elo：39.56[32.40, 47.02]  [font=&quot]显著性水平（LOS）：100.00  [font=&quot]LLR：3.25[-2.94, 2.94][font=&quot]时间控制：60+0.6  [font=&quot]总场次/胜/平/负：2926 / 783 / 1601 / 542  [font=&quot]PTNML：1 / 167 / 893 / 394 / 8  [font=&quot]胜率：54.12%  [font=&quot]Elo：28.53[22.92, 34.31]  [font=&quot]显著性水平（LOS）：100.00  [font=&quot]LLR：3.00[-2.94, 2.94][font=&quot]时间控制：180+1.8  [font=&quot]总场次/胜/平/负：3066 / 807 / 1671 / 588[font=&quot]PTNML：0 / 155 / 1005 / 372 / 1  [font=&quot]胜率：53.57%  [font=&quot]ELO：24.72[19.70, 29.88]  [font=&quot]LOS：100.00  [font=&quot]LLR：2.96[-2.94, 2.94][font=&quot]Pikafish 继续引领引擎实力的标杆。在面对最强对手时，它始终稳居引擎锦标赛榜首，持续主导着整个领域。[font=&quot]更新亮点[font=&quot]通用二进制文件  [font=&quot]我们已将发布版本转为通用二进制格式，简化了下载流程。这些通用二进制文件可自动检测您CPU的特性并运行最优代码，无需再手动选择AVX2、AVX-512等选项。[font=&quot]升级的NNUE架构与训练  [font=&quot]本次发布升级了NNUE网络架构，通过移除冗余的威胁特征，有效减小了二进制文件体积。[font=&quot]训练过程进一步优化，引入了量化感知训练（QAT）等新技术，并对参数进行了精细调整。这些技术已应用于数百亿个训练局面，所有局面均使用强大的Px0网络进行了持续重新评分。[font=&quot]扩展的平台支持[font=&quot]我们已为 RISC-V（RVV）和 LoongArch（LSX/LASX）添加了原生支持，同时支持 1GB Linux 巨型页以及 WebAssembly 目标平台。此外，针对 Linux、macOS 和 BSD 的共享内存实现也已全面重构。[font=&quot]严格的位置验证  [font=&quot]我们对棋盘位置、FEN 字符串和 UCI 命令实施了更严格的验证。引擎现在会输出“info string CRITICAL ERROR”，随后显示确切的命令及失败原因，并立即终止进程。一个良好的图形界面将确保您不会遇到此类错误。[font=&quot]谢谢你[font=&quot]Pikafish 基于 Stockfish 构建，而 Stockfish 是一个由活跃的爱好者社区支持的项目（感谢每一位贡献者！）。他们贡献自己的专业知识、时间和资源，共同打造了一款强大、广泛可用且开源免费的国际象棋引擎。[font=&quot]我们衷心感谢为我们的 GitHub 项目点亮了 2000 颗星标。感谢你们的支持与鼓励——你们的认可对我们意义重大。程序员们可以通过直接参与 Stockfish（C++）、Fishtest（HTML、CSS、JavaScript 和 Python）、训练器 nnue-pytorch（C++ 和 Python），或网站开发（HTML、CSS/SCSS 和 JavaScript）等方式为上游项目做出贡献。[font=&quot]Pikafish 团队
https://github.com/official-pikafish/Pikafish/releases/download/Pikafish-2026-09-06/Pikafish.2026-09-06.7z

祝贺皮卡鱼团队，感谢分享优秀作品，期待正式版本

祝贺皮卡鱼团队，感谢分享优秀作品，期待正式版本

   祝贺皮卡鱼团队

本次新版本的 Elo 增益高达 24 点，赢得的对局对数超过输掉的两倍。

本次发布升级了NNUE网络架构，通过移除冗余的威胁特征，有效减小了二进制文件体积。训练过程进一步优化，引入了量化感知训练（QAT）等新技术，并对参数进行了精细调整。

待我下载来瞧瞧 。。。。   看来是一个 ELO 小跳跃版本  

祝贺皮卡鱼团队，感谢分享优秀作品，期待正式版本，另，鲨鱼界面里面，fen命令选项要把fen改为position fen显示才正常。

祝贺皮卡鱼团队，感谢分享优秀作品。

原文：
Today, we have the pleasure of announcing Pikafish 2026-09-06. As always, you can freely download it at GitHub Releases and use it as a drop-in replacement in the GUI of your choice to benefit from stronger play and more accurate analysis.

Whether you can spare hours or days of CPU time, your help matters for the ongoing development of Stockfish, which also strengthens the foundation on which Pikafish is built. Find out how you can contribute at stockfishchess.org/get-involved. Join their Discord server to get in touch with the community of developers and users of the project!

Quality of Xiangqi Play
In tests against Pikafish 2026-01-02, this new release brings an Elo gain of up to 24 points, and wins more than twice as many game pairs as it loses.

Book: winrate 65_85

TC: 10+0.1
Total/Win/Draw/Lose: 2766 / 830 / 1421 / 515
PTNML: 7 / 193 / 688 / 468 / 27
WinRate: 55.69%
ELO: 39.56[32.40, 47.02]
LOS: 100.00
LLR: 3.25[-2.94, 2.94]

TC: 60+0.6
Total/Win/Draw/Lose: 2926 / 783 / 1601 / 542
PTNML: 1 / 167 / 893 / 394 / 8
WinRate: 54.12%
ELO: 28.53[22.92, 34.31]
LOS: 100.00
LLR: 3.00[-2.94, 2.94]

TC: 180+1.8
Total/Win/Draw/Lose: 3066 / 807 / 1671 / 588
PTNML: 0 / 155 / 1005 / 372 / 1
WinRate: 53.57%
ELO: 24.72[19.70, 29.88]
LOS: 100.00
LLR: 2.96[-2.94, 2.94]

Pikafish continues to set the standard for engine strength. Against the strongest competition, it consistently secures the top spot in engine championships, continuing to dominate the field.

Update Highlights
Universal Binaries
We have transitioned to universal binaries for our releases, simplifying the download process. These universal binaries automatically detect the features of your CPU and run the optimal code, eliminating the need to manually choose between AVX2, AVX-512, etc.

Upgraded NNUE Architecture and Training
This release upgrades the NNUE network architecture, reducing binary size by removing redundant threat features.

The training process has been further improved with the introduction of new techniques, such as Quantization-Aware Training (QAT), and further parameter tweaks. These techniques have been applied to hundreds of billions of training positions, all of which have been consistently rescored using a strong Px0 net.

Expanded Platform Support
We have added native support for RISC-V (RVV) and LoongArch (LSX/LASX), 1GB Linux huge pages, as well as WebAssembly targets. The shared-memory implementation for Linux, macOS, and BSD was also overhauled.

Strict Position Validation
We have implemented stricter validation for board positions, FEN strings, and UCI commands. The engine will now output an followed by the exact command and the reason it failed, and then immediately terminate the process. A good GUI will ensure you never encounter these errors.info string CRITICAL ERROR

Thank You
Pikafish is built on top of Stockfish, a project supported by a thriving community of enthusiasts (thanks to everybody!) who contribute their expertise, time, and resources to build a free and open-source chess engine that is robust, widely available, and very strong.

We would like to express our gratitude for the 2k stars that light up our GitHub project. Thank you for your support and encouragement – your recognition means a lot to us. Programmers can contribute to the upstream project either directly to Stockfish (C++), to Fishtest (HTML, CSS, JavaScript, and Python), to their trainer nnue-pytorch (C++ and Python), or to their website (HTML, CSS/SCSS, and JavaScript).

The Pikafish team