Skip to content
Tech News
← Back to articles

Stockfish 19

read original get DGT Centaur Electronic Chess Board → more articles
Why This Matters

Stockfish 19 raises the bar for the world's strongest open-source chess engine, adding up to 44 Elo over version 18 through a new SFNNv16 neural network architecture and modern training techniques like quantization-aware training. Beyond chess, the release shows how community-driven, distributed-compute projects continue to push efficient neural network design, and its new universal binaries and RISC-V/LoongArch support make it easier to run across hardware.

Key Takeaways
Worth a Look

DGT Centaur Electronic Chess Board — If Stockfish 19's stronger analysis has you itching to play over the board, the DGT Centaur pairs a real weighted piece set with a built-in adaptive engine and LED-guided move display. It's a great way to train on physical wood-feel pieces while still getting engine-grade feedback on your games.

See DGT Centaur Electronic Chess Board on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.

Today, we have the pleasure of announcing Stockfish 19, a new major release. As always, you can freely download it at stockfishchess.org/download 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. Find out how you can contribute at stockfishchess.org/get-involved. Join our Discord server to get in touch with the community of developers and users of the project!

Quality of Chess Play

In tests against Stockfish 18, this new release brings an Elo gain of up to 44 points, and wins more than three times as many game pairs as it loses.

Stockfish 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.

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 introduces the SFNNv16 network architecture, reducing binary size by removing redundant threat features while increasing strength by introducing new pawn-pair features. The secondary neural network, introduced in Stockfish 16.1, has been retired, enhancing strength in positions where the small net previously underperformed.

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 Leela net.

... continue reading