Reka AI trained Flash 3.1 using math answers and executed code as checkable feedback. Its training account describes filtering problems that cannot be verified reliably, converting multiple-choice questions into answer-entry questions to discourage shortcuts, and testing generated programs against several examples. The aim is to reward solutions that survive a concrete correctness check. Training method.
The companion release announcement reports a 10-point improvement over Flash 3 on LiveCodeBench v5, a coding benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition →. That provides a reported result for the upgraded training system, although it does not isolate the contribution of each training change. Reported comparison.
Terms in this piece · Glossary
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
