Evolution of Colibri2

The releases of Colibri2 on the benchmarks that guide its development, run on the same cluster with 60 seconds and 4 GB per goal:

ReleaseDate Why3 goals proved SMT-LIB solved Time on the 40089 Why3 goals all prove Wrong answers
0.4 2023-02-06 41013 84.6% 887 28.4% 53505 s 4
0.5 2026-02-25 45070 93.0% 1881 60.2% 13805 s 0
0.6 2026-06-22 45140 93.1% 1876 60.0% 14712 s 0
0.7 2026-11-16 45912 94.7% 1959 62.7% 14261 s 0

Wrong answers: sat on a goal known to be valid, not counted as solved. The four of 0.4 are on bit-vectors.

Floating-point arithmetic

The floating-point problems of SMT-LIB, with the same limits:

ReleaseDate Solved without quantifiers Solved with quantifiers Time on the 57526 problems without quantifiers all solve Wrong answers
0.5 2026-02-25 57565 99.5% 723 82.9% 2988 s 0
0.6 2026-06-22 57564 99.5% 722 82.8% 3789 s 0
0.7 2026-11-16 57624 99.6% 739 84.7% 5862 s 0

Wrong answers: an answer that contradicts the known status of the problem. 0.4 is not in this table: it was far behind on the floating-point problems.

The benchmarks and the scripts: colibri-regression.