CodeGolf Bench - C++ Pass@1: leaderboard

CodeGolf Bench C++ Pass@1 asks LLMs to solve code.golf problems in C++ with as few characters as possible and reports the share of problems whose first solution passes all hidden platform tests, estimated from 10 samples (CodeGolf Bench). Higher is better.

Metric: Pass@1 (%): share of problems whose C++ solution passes every hidden test, estimated from 10 samples per problem; 115 code.golf problems (holes), solutions checked against the platform's hidden tests through its API; golfing prompt, temperature 0.2, top-p 0.95, up to 32,768 output tokens, 10 samples per problem; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Flash (Preview 04-17)43.24#453
2DeepSeek R139.13#482
3Qwen 3 235B A22B37.02#628
4Gemini 2.5 Pro (Preview 03-25)25.2#241
5Gemma 3 27B (IT)24.49#1324
6DeepSeek R1 Distill Llama 70B23.41#1669
7Llama 4 Maverick20#1110
8DeepSeek V3 (0324)19.46#715
9Llama 4 Scout6#1818

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=codegolf-bench-c-plus-plus-pass-1 · How It Works · Data refreshed daily, snapshot 2026-11-21.