TIS-100 Bench
by Wincon
What is TIS-100 Bench? TIS-100 Bench is a coding eval made by Wincon. A benchmark that has models write and optimize assembly-style programs for the programming puzzle game TIS-100, where every solution is scored on cycles, nodes and instructions. Models get the manual, visible test inputs and game feedback through text tools, but not the community leaderboard, and have one scored attempt per puzzle across 55 puzzles (the 47 in the game plus eight community tournament puzzles) with 10 to 45 minutes each. A result only counts as a new frontier point if it beats or trades off better than every community solution (leaderboard frozen 15 Oct 2026) and passes 100,000 tests in an independent simulator. In the 3 Nov 2026 snapshot, 16 models found 133 new frontier points on 18 puzzles; Opus 5.5 and GPT-6 Astra led and the authors say the data doesn't separate them. The models' programs are published under the MIT License.
Editor note
A benchmark that has models write and optimize assembly-style programs for the programming puzzle game TIS-100, where every solution is scored on cycles, nodes and instructions. Models get the manual, visible test inputs and game feedback through text tools, but not the community leaderboard, and have one scored attempt per puzzle across 55 puzzles (the 47 in the game plus eight community tournament puzzles) with 10 to 45 minutes each. A result only counts as a new frontier point if it beats or trades off better than every community solution (leaderboard frozen 15 Oct 2026) and passes 100,000 tests in an independent simulator. In the 3 Nov 2026 snapshot, 16 models found 133 new frontier points on 18 puzzles; Opus 5.5 and GPT-6 Astra led and the authors say the data doesn't separate them. The models' programs are published under the MIT License.
- What it tests
- Whether models can not only solve tight low-level programming puzzles but keep optimizing them past human community records.
- Original source
- tis-100.wincon.dev/
- Last updated
- Snapshot 2026-11-03-v5
- License
- Not checked yet — see source
Related coding evals
- SWE-bench — Fixing real bugs in real repositories, judged by unit tests.
- HumanEval — Writing a short correct function from a docstring.
- LiveCodeBench — Code generation, self-repair and test prediction on fresh problems.
- Aider leaderboards (Polyglot) — Practical multi-language code editing.
Discussion
No one has discussed TIS-100 Bench yet. Have you run it? Share what it caught, what it missed or how you set it up.