New engine releases & news H1 2026

[Graham Banks]

I would prefer to go with the current 4.1 for the CCRL rating.

In the next weeks/months I will come back with my own net and a big update of the engine, but for now I would prefer for you to test the current version, if possible.

Thank you, let me know if you need any details.

Only if you use your own net with this version (4.1).
Otherwise, we'll wait.

[NathanDrake]

I would prefer to go with the current 4.1 for the CCRL rating.

In the next weeks/months I will come back with my own net and a big update of the engine, but for now I would prefer for you to test the current version, if possible.

Thank you, let me know if you need any details.

Only if you use your own net with this version (4.1).
Otherwise, we'll wait.

Ok, I will add two more .zip files to the release with my own net. The engine uses a small net as well (also from SF). Do I need to remove that one too? Or is it sufficient to have my own net for the big net?

[DreamerExx]

Ember 1.0.0
Strength: Around 2700
https://github.com/ExxDreamerCode/Ember

[Enderjed]

Alecto
Made collaboratively by "Amolyte" (Minnesota) and Jedidiah F. Sessions (England)
https://www.dropbox.com/scl/fo/34bvfbya ... kk3d6&dl=0
Source code and logo is included, although has only been compiled for a Windows .exe

Much like Valiant, which it is derived from (although is written here in Rust), it contains a "Worstfish" toggle that inverts the evaluation, making the engine search for the worst possible moves instead of best.

[supernova]

Hey guys,

PrechessUCI 1.33 is out. Fun fact: this originally started as a very rudimentary engine that could only play with the Black pieces, but that has been fully corrected. It now plays both sides flawlessly. I spent the last few weeks completely rewriting the internals, and it's running great.

The main upgrade is speed. The Quiescence Search was wasting a ton of time generating quiet moves just to throw them away. I wrote a strict capture-only generator that skips the overhead, bumping the base speed from ~860k up to 1.1 Million NPS.

I also built a native ML tuner in Ada and ran a multithreaded coordinate-descent over 725,000 positions. I hardcoded all 786 tuned evaluation weights (piece-square tables, penalties, etc.) straight into the compiler as constants. It plays much better positionally now without losing any speed.

For the purists, José's original legacy search from Prechess 0.7.8 is still in there as a UCI option, just heavily optimized.

Two quick limitations: the search is strictly single-threaded, and the Hash size is locked at 33MB. The old code relies heavily on global variables for the board state, so spinning up multiple threads for Lazy SMP would just cause instant race conditions and crashes. Fixing that requires a full stateless rewrite of the legacy code, which I haven't done yet. Changing 'Threads' or 'Hash' in your GUI won't do anything.

Distribution follow the MIT License. The zip has the Ada source code and a highly optimized Windows 64-bit executable (prechess_uci64.exe).

If someone want to give it a run in Arena or CuteChess, let me know if it crashes or blunders.

Cheers, Arthur (supernova).

https://drive.google.com/file/d/16fO8Zd ... drive_link

[Graham Banks]

I would prefer to go with the current 4.1 for the CCRL rating.

In the next weeks/months I will come back with my own net and a big update of the engine, but for now I would prefer for you to test the current version, if possible.

Thank you, let me know if you need any details.

Only if you use your own net with this version (4.1).
Otherwise, we'll wait.

Ok, I will add two more .zip files to the release with my own net. The engine uses a small net as well (also from SF). Do I need to remove that one too? Or is it sufficient to have my own net for the big net?

Best to remove SF nets.

[Graham Banks]

New experimental version of Firefly - https://github.com/r28666/FireFlyChessBinaries

[Elorejano]

Hi, Graham.
FireFly (if i remember ) needs some files that ar absent. Like an ini file, openings books etc
Do you have those ones?

[Graham Banks]

Hi, Graham.
FireFly (if i remember ) needs some files that ar absent. Like an ini file, openings books etc
Do you have those ones?

It is a UCI engine, so the options will become available when loading it.

[NathanDrake]

I would prefer to go with the current 4.1 for the CCRL rating.

In the next weeks/months I will come back with my own net and a big update of the engine, but for now I would prefer for you to test the current version, if possible.

Thank you, let me know if you need any details.

Only if you use your own net with this version (4.1).
Otherwise, we'll wait.

Ok, I will add two more .zip files to the release with my own net. The engine uses a small net as well (also from SF). Do I need to remove that one too? Or is it sufficient to have my own net for the big net?

Best to remove SF nets.

Done. I've uploaded a build with my own net and no Stockfish nets at all, exactly as you suggested.

A quick note on the label: I tagged it 4.2 because the network changed and some tuned parameters, but it's the same engine and search as 4.1. Only the net is different. The big net is now my own rubicon net (trained from scratch), and the SF small net is gone too: the engine runs big-net only.
https://github.com/Tors3/Triumviratus/releases/tag/v4.2

This is my first own net, so it's a little weaker than the SF-net version, but it's a real own-lineage milestone for the engine. I expect to come back with a stronger own net in the coming months; for now this one is good to go.

Thanks for testing, and let me know if you need anything else!