Fable 5.2 chess 24hrs: 3542 Elo, written from scratch by an AI in 24 hours

[Steve Maughan]

The newest engine from my 24-Hour Chess Engine Benchmark is out, and it is a big one.

Fable 5.2 chess 24hrs was designed, written, trained and tuned by Anthropic's Fable 5.2 model in a single autonomous 24-hour session. It started with an empty folder and a C++ compiler. Nobody answered questions, and no existing engine code, network or book was allowed. It did not even look anything up online. It wrote its own NNUE trainer and trained the network on its own self-play games along the way.

Rating: 3542 ± 15 on the CCRL Blitz scale (2,220 games at 10s+0.1s, single thread, against ten CCRL-rated engines and three of the other AI engines).

Why I'm excited about this one:

• +81 Elo over Opus 5.5 (3461), which had only just set the record with a 207 Elo jump over Fable 5.1.
• +288 Elo over Fable 5.1 (3254). That is the gain from one point release of the same model line, under the same rules and the same 24 hours.
• It is the first engine in this series that I'd say has an attractive playing style. It doesn't just win on tactics; the games are a pleasure to watch, and it feels like a properly strong engine.
• It is about 260 Elo behind Stockfish 19, which is 3803 on the current CCRL Blitz list (and that is Stockfish on 8 CPUs against this engine's single thread). For one day's unassisted work, I find that remarkable.

Some results from the rating run: level with Patricia 5.0 (49.8%), 55.5% against Sirius 9.0, 57.8% against Carp 3.0.1 and 79% against Stash 37. It had no time losses, crashes or illegal moves in all 2,220 games.

What's inside:

• PEXT bitboards, about 2,200 lines of C++
• PVS with null move, LMR, singular extensions, ProbCut, and history and correction-history tables
• NNUE: (768x5 -> 256)x2 -> 1x8, SCReLU, horizontally mirrored, trained from zero on about 175M self-play positions (CPU only, no ML framework)
• 55 search parameters tuned by SPSA during the session

Download (Windows x64, needs AVX2 and BMI2):
https://github.com/stevemaughan/fable-5 ... s/tag/v1.0

Source, the model's own write-up and its hour-by-hour log:
https://github.com/stevemaughan/fable-5.2-chess-24hrs

The benchmark rules and template, if you want to try another model:
https://github.com/stevemaughan/chess-engine-benchmark

I'd love to hear what you think of its play, and independent test results are very welcome.

Steve

[Kaltago]

Thank you for sharing your adventures with us. I find it fascinating that such strong engines can be created so quickly. Each of your engines so far have been written for 1 core. Any chance you would add multiple core support?

Thank you. I look forward to your continued adventures!

[Steve Maughan]

Initially I didn’t want engines to waist time on pondering and multi-core support. As the AI has developed these seem like trivial additions. Maybe I’ll add those to a new benchmark. Thanks for the interest!

— Steve

[Modern Times]

I'm amazed at what is possible with AI. In future we may not see any authors creating engines the traditional way anymore. Then it will be all about the skill and capabilities of the AI and LLM, the ability of a human "director" to guide them accordingly and then the computational power residing in data centres to produce the result for review. All very impressive and we can't stand in the way of progress, but computer chess then becomes something totally different from what it is today. A different sort of competition altogether. One that I understand but won't be a part of when it gets to that point.

[madushan1996]

Here, before some anti - AI person screams, "AI slop!"

Actually, any frontier LLM today is better at coding than any human on the planet. So, if someone calls AI generated code "AI slop," then human written code would be "slop pro max."

The process of creating a strong chess engine has shifted from being primarily a coding challenge to a machine learning challenge.Anyone can run an agent loop powered by top tier model for days or weeks continuously to optimize their engine. I've seen the codebases of many chess engines where AI was clearly used to write code or later optimizations , but the authors haven't publicly acknowledged it.

The following are the factors I believe will contribute to the strength of chess engines in the future:

1. The author's access to computational resources
2. The quality of the training dataset
3. The volume of the training dataset
4. The author's understanding of machine learning and deep learning principles
5. The author's overall understanding of how the engine operates
6. The author's ability to understand AI generated output
7. The author's ability to identify and understand bugs quickly
8. The author's creativity and ability to think outside the box and ability come up with novel architectures.

In the next few months, I believe we'll see more and more chess engines developed by individuals or much smaller teams going to be a challenge to Stockfish, Reckless etc will eventually overthrow them.

[op12no2]

I would hope/predict however that the human-competitive-coding orientated authors would fight to keep a sub-space of artisans. Yes it's not completely policeable, but a signature would go a long way. There is already a rating list emerging based on artisan principles.

https://cope-chess.live/

But I also find Steve's experiments fascinating...

[mar]

...
Actually, any frontier LLM today is better at coding than any human on the planet. So, if someone calls AI generated code "AI slop," then human written code would be "slop pro max."
...

... and then you woke up from a wet dream

[madushan1996]

...
Actually, any frontier LLM today is better at coding than any human on the planet. So, if someone calls AI generated code "AI slop," then human written code would be "slop pro max."
...

... and then you woke up from a wet dream

Keep laughing and stay ignorant inside your tiny bubble. you'll have nothing left to say!

[Ipmanchess]

Fable 5.2 will be added here in this new AI-list

https://ipmanchess.yolasite.com/r9-7945 ... ngines.php

Ipman.

[Uri Blass]

How much did you pay to get the chess engine?

I asked google
how much do you need to pay in order to use Fable5.2 to work for 24 hours to build chess engine

AI replied:

Running Claude Fable 5.2 continuously for 24 hours to build a chess engine will likely cost between $300 and $900+ via the API, as standard consumer subscriptions enforce strict usage caps that will pause a marathon session.
Because Fable 5.2 is billed by token usage rather than by the hour, your final bill depends heavily on how aggressively your coding agent loops, tests, and refactors code.

Edit:After asking a different question the claim is that the price is clearly smaller and I get replies that using claud pro is only 20$ for month and you can use tokens to do the task cheaper so I am not sure about the price.

I am also not sure if using claud pro is the best AI that we have to write chess applications.
I use gemini and not claud pro and I would like to know if generally claud pro is better for writing chess applications.