Aggressive NNUE chess engine · plays like Tal
A modern neural-network (NNUE) chess engine built to attack — bold, sacrificial, alive — in the romantic style of World Champion Mikhail Tal, and one of the most aggressive engines around. Created by Chris Whittington at Oxford Softworks in the 1990s, and revived since 2023 with veteran programmer Ed Schröder of Rebel: Chris on the engine code, Ed on the neural networks and the techniques that shape its style of play.
One dramatic result is Chess System Tal Extreme, in the tradition of the old geniuses — Morphy, Steinitz, Lasker, Anderssen — which tops both playing-style rating lists. Wilder still is the Absurd Engine with some example wild games.
Free to download — the latest Chess System Tal engine and several EPD test suites: some for exhaustive testing of PERFT, and others for use as randomised opening positions for testing with various common chess-odds games.
The modern NNUE engine (UCI protocol), E1019 net — tuned for raw strength. Pick the build matching your CPU: most machines want AVX2; choose AVX512 only if your CPU supports it (about 20% faster, though it varies by machine). ~40 MB each. Old PC? The Legacy build (scalar — no AVX) runs on any 64-bit CPU.
📥 Downloads to date — AVX2: — · AVX512: — · scalar: —
All releases & notes on GitHub →The same engine with the E1162-EAS net — tuned for Tal-style attacking play (EAS: bold and sacrificial). Pick the build matching your CPU: most machines want AVX2; choose AVX512 only if your CPU supports it (about 20% faster, though it varies by machine). ~40 MB each. Old PC? The Legacy build (scalar — no AVX) runs on any 64-bit CPU.
📥 Downloads to date — AVX2: — · AVX512: — · scalar: —
All releases & notes on GitHub →The most aggressive competitive build, playing in the tradition of the old masters — Morphy, Steinitz, Lasker, Anderssen — and topping both playing-style rating lists. Now in AVX2 and AVX512 — pick the build matching your CPU (AVX2 runs everywhere). ~33 MB each. Old PC? The Legacy build (scalar — no AVX) runs on any 64-bit CPU.
📥 Downloads to date — AVX2: — · AVX512: — · scalar: —
More on Chess System Tal Extreme →An even wilder, crazier attacking style — bold sacrifices taken to the extreme. Now in AVX2 and AVX512 — pick the build matching your CPU (AVX2 runs everywhere). ~33 MB each. Old PC? The Legacy build (scalar — no AVX) runs on any 64-bit CPU.
📥 Downloads to date — AVX2: — · AVX512: — · scalar: —
More on the Absurd Engine →Various EPD test suites — position sets for testing and analysing chess engines. 59 files (PERFT, standard suites and chess-odds positions), subdirectories preserved.
📥 — downloads to date
Browse on GitHub →The Universal Chess Interface — the command-line protocol between chess engines and GUIs (original specification by Stefan Meyer-Kahlen).
📥 — downloads to date
View on GitHub →Polyglot book binary files for chess engines that support the Polyglot book format — a selection of opening books to choose from.
📥 — book downloads to date
Browse on GitHub →One preference runs through all of it: knowledge over brute force. Where most of computer chess leaned on ever-faster, ever-wider search, the bet here was on encoding real chess understanding instead.
In short, a counter-current in computer chess: the idea that chess intelligence is best expressed through structured knowledge and creative heuristics — not raw search power alone.
Four decades of computer chess, 1982 to today.
An early commercial chess program — the start of a long road.
A popular commercial release from Oxford Softworks.
Continuing the Oxford Softworks chess line.
A full chess package for MS-DOS (Amiga in 1994), with its own graphical interface and 2D/3D boards. Preserved today on the Internet Archive.
View on Internet Archive →The original release and its sequel — Oxford Softworks' most daring engines, and the ones that made the name.
A modern revival: a UCI engine in C++ (with Ed Schröder) using a neural-network (NNUE) evaluation, yet still tuned for Tal-style aggression. Rated around 2914 blitz and ranked among the top publicly available engines, with a "Learn Assist" feature that steers the search toward preferred lines.
View on GitHub →Currently in development.
British chess and games programmer, publisher and entrepreneur, founded Oxford Softworks in Burford, Oxfordshire in the mid-1980s and spent two decades building chess engines — from SuperChess (1982) to the Complete Chess System and, above all, Chess System Tal, designed to play in the romantic, sacrificial style of World Champion Mikhail Tal. Oxford Softworks became a software development, licensing and publishing house specialising in strategy games — Bridge, Go, Shogi and others — and was sold to venture capitalists in May 2000; the chess engine never retired, and Chess System Tal lives on today as a modern neural-network (NNUE) engine.
After the sale, Chris retired (sort of) to a riverside house near Pershore, Worcestershire and turned to other pursuits: vegetable patch, beekeeping, alpacas, sheep, pigs, cows and river boating, programming remaining a part-time pursuit including a neural-net Backgammon project and a stock-market analysis tool (these never work btw — there is no way a retail investor can get ahead of the market sufficiently to overcome trading friction). For a time he also served as Chairman of the Parish Council, handing over to the capable hands of Bob Annis in 2009.
In 2010 was the move to a farmhouse in south-west France — bees, honey, vegetable patch, carpentry workshop, generally things rural — until there came AlphaZero and AlphaGo, the big shock from DeepMind that neural nets could do what was thought impossible: play chess above Grandmaster level. So, back to AI and machine learning — that was some catching up to do, discovering the strange thing called NNUE and, of course, what better way to learn than to do it oneself, or, as it turned out, twoselves.
Thus was Chess System Tal NNUE, and the development partnership with veteran chess programmer Ed Schröder of Rebel fame, born. Chris worked on the engine code; Ed was responsible for the neural networks, and especially for the techniques that dramatically shape the engine's style of play — one dramatic result being Chess System Tal Extreme, in the tradition of the old geniuses of yesteryear such as Morphy, Steinitz, Lasker and Anderssen, which tops both playing-style rating lists. Wilder still is the Absurd Engine.
Since 2025 Chris turned digital-nomad-adventurer, keeping the French farmhouse but much of the year now spent in South-East Asia and the Caucasus. With a couple of laptops and an internet connection, the development of a new Chess System Tal 3 has begun. All good and promising so far. News as it develops…
Chess System Tal 3 judges positions with a neural network that holds no piece-value table at all — so we reverse-engineered one. By regressing millions of the engine's own evaluations we recovered what each piece is truly worth to it, how those values slide from a crowded opening to a bare endgame, and the piece-square maps it taught itself with nothing hand-coded: knights that crave outposts, a king that flips from cowering in the corner to marching up the board, and a queen worth nearly nine pawns when there's a full board to attack.
A growing collection of tools, formats and resources for chess-engine authors — the practical building blocks gathered over four decades of writing chess software. From the UCI protocol and Polyglot opening-book format to EPD test suites for PERFT and engine analysis, these are the bits and pieces that make developing and testing a chess engine that much easier. More to come.
Various EPD test suites — position sets for testing and analysing chess engines. 59 files (PERFT, standard suites and chess-odds positions), subdirectories preserved.
📥 — downloads to date
Browse on GitHub →The Universal Chess Interface — the command-line protocol between chess engines and GUIs (original specification by Stefan Meyer-Kahlen).
📥 — downloads to date
View on GitHub →Polyglot book binary files for chess engines that support the Polyglot book format — a selection of opening books to choose from.
Browse on GitHub →📥 — book downloads to date
A Python research engine that trains LSTM models to forecast forward returns across a basket of ETFs (gold miners, energy and the Nasdaq-100), backtests each with realistic transaction costs, and runs a confidence-ranked rotation strategy that shifts capital toward the strongest signals — falling back to cash when none are convincing. Built with PyTorch, walk-forward cross-validation, and a fully reproducible train → backtest → rotate → live-signal pipeline. Best used from a Claude Code interactive window — then you can speak to the engine in English rather than computer-ese.
Disclaimer: a personal research project — not investment advice. Backtested results are historical simulations and do not guarantee future performance. Never risk real money based on scenario back-tests. Always trial out trading strategies in real time using paper, not real money. What may work in one trading regime/time frame may not work in another. In my humble opinion it is extremely difficult for amateur investors to get ahead of the professionals, even using AI and neural networks. It is extremely easy to lose all your money. Slippage and friction don't help. Use absolutely at your own risk.
📥 — source downloads to date
A companion project: a forensic read of chess-engine source code, judging each repository on two axes — was it written by a human or generated by AI, and how close does it sit to Stockfish? Every engine is cloned, its git history and code scanned for the tell-tale fingerprints, then contrasted with Stockfish.
So far, more than fifty engines. Fourteen newly-released ones get full individual write-ups — among them two that proved AI-generated (one a careless paste that still had the AI's own citation markers left in the comments; one assembled by a coding agent) and several that are genuinely AI-assisted. On top of that, a clone-and-scan sweep of the entire CCRL 40/40 top 50 — the strongest engines in the world. The verdict from that elite tier is a story in itself: no AI-generated engines, exactly one quietly AI-assisted (Arasan, human since 1994), and not a single actual Stockfish fork — they have all simply converged, independently, on the same winning design.
Forensic source reads across 14 engines — from a repo with leftover AI citation markers, through an agent-built one, to expert human projects — each placed on an authorship spectrum from AI-generated, through AI-assisted, to fully hand-written.
Read the authorship analysis →A second, neutral spectrum: how close is each engine to Stockfish? From independent designs — their own board representation, hand-crafted evaluation, unusual languages — through to engines built directly on Stockfish's own search and network.
See the Stockfish-similarity spectrum →The same tests run across the strongest engines in the world. The result: no AI-generated engines, a near-total convergence on the Stockfish recipe — and one surprise: Arasan, human since 1994, now AI-assisted.
See the CCRL Top 50 survey →For anything about Chess System Tal, Oxford Softworks, or computer chess, the engine lives on GitHub. The author can often be found at the TalkChess forum.
Chess System Tal on GitHub