AetherCode: leaderboard

Each model writes solutions to 456 problems from 2024 and January-May 2025 olympiad and ICPC contests, sampled four times per problem with up to 32,768 output tokens; the benchmark's authors at ByteDance Seed ran every model, including their own Seed-1.6-thinking. The score is Pass@1, the average share of samples that pass test suites validated against more than 30,000 human submissions; the problems come from premier contests rather than Codeforces or LeetCode as in LiveCodeBench.

Metric: Pass@1 (%). Source: huggingface.co. Saturation forecast: Around February 2027. 9 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1O4 MiniHigh35.5#367
2Seed-1.6-thinking-071526.6#492
3DeepSeek R1 052822.3#431
4Qwen 3 235B A22BThinking17.6#631
5Qwen 3 32BThinking16.3#965
6GPT-4.110.5#505
7Qwen 3 8BThinking10#1868
8DeepSeek V3 (0324)8.5#722
9Qwen 3 Coder 480B A35B Instruct7.7#653

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=aethercode · How It Works · Data refreshed daily, snapshot 2026-11-17.