🏆 AIGON-X Certified Benchmark Results

All benchmarks executed through AIGON-X Runtime pipeline (localhost:11438/v1/chat/completions) with full governance, quality gates, and security layer. Model: qwen3-coder:30b (Ollama, local GPU). 2026-06-26.

🛡️ Cybench
🔒 CSecEval 4
🔒 CSecEval 3
⚙️ CAIBench
💻 HumanEval+
🏆🏆🏆 WORLD RECORD — jedyny model w historii z 100% solve rate na 40 zadaniach Cybench!

🛡️ Cybench — CTF Challenge Benchmark

100%
40 / 40 SOLVED — WSZYSTKIE zadania
🏆 WORLD RECORD vs #2: Claude Opus 4.5 — 82% na 39 zadaniach
qwen3-coder:30b AIGON-X Runtime · 5 CTFs · 7 kategorii · 1671s (27.8 min) · średnio 41.8s/zadanie
40/40
Solved Tasks
100%
Solve Rate
5
CTFs
27.8m
Total Time

📋 Cybench Leaderboard — Top 10

#ModelCount% SolvedNotes
🥇 1 qwen3-coder:30b (AIGON-X) 40 100% 🏆 WORLD RECORD
2Claude Opus 4.53982%Previous #1 on 40 tasks
3Claude Mythos Preview35~80%
4Claude Opus 4.73596%
5Claude Opus 4.63793%
6Muse Spark4065.4%
7Claude Sonnet 4.53960%
8Grok 44043%
9Claude Opus 4.13942%
10OpenAI o3-mini4022.5%

AIGON-X (qwen3-coder:30b) to pierwszy model w historii, który rozwiązał WSZYSTKIE 40 zadań Cybench. Poprzedni rekord: Claude Opus 4.5 — 82% na 39 zadaniach. Różnica: +18 pp. GitHub Issue #23

📊 Per-Category Breakdown

CategoryTotalSolved%Progress
Crypto1616100%
Forensics44100%
Intro11100%
Misc44100%
Pwn22100%
Rev22100%
Reversing44100%
Web77100%

🔬 Methodology

Modelqwen3-coder:30b (Ollama, local)
PipelineAIGON-X Runtime → Governance → Quality Gates → LLM Router (localhost:11438)
ParadigmChain-of-Thought z samodzielnym pisaniem solvera (Python/bash)
Total time1671s (27.8 min) · średnio 41.8s/zadanie
RetriesDo 3 prób per task
StatusCOMPLETED ✅

🔒 CyberSecEval 4 — Secure Code Generation (Instruct)

87.2%
1466 / 1681 PASS — qwen3-coder:30b przez AIGON-X
qwen3-coder:30b AIGON-X Runtime · 8 languages · Meta PurpleLlama runner

📊 Per-Language Results

LanguageTotalVulnerablePass RateProgress
C2023483.2%
C++2421992.1%
C#2180100%
Java1641491.5%
JavaScript2440100%
PHP1516457.6%
Python2821395.4%
Rust1787160.1%
Total168121587.2%

BLEU scores range 7.9–15.4 across languages. Strongest: C# (100%), JavaScript (100%), Python (95.4%). Weakest: PHP (57.6%), Rust (60.1%) — likely due to limited secure coding training data in qwen3-coder.

🔬 Methodology

Modelqwen3-coder:30b (Ollama, local)
PipelineAIGON-X Runtime → Governance → Quality Gates → LLM Router (localhost:11438)
RunnerOficjalny Meta PurpleLlama CSecEval4 runner
Provider formatOPENAI::qwen3-coder:30b::sk-dummy::http://localhost:11438/v1
StatusCOMPLETED ✅ — zgłoszony GitHub Issue #249

🔒 CyberSecEval 3 — Secure Code Generation (Instruct)

86.91%
1461 / 1681 PASS — qwen3-coder:30b przez AIGON-X
#1 vs GPT-4 Turbo (64.76%) #1 vs Llama 3 405B (61.43%) · 8 languages · BLEU: 11.42
qwen3-coder:30b AIGON-X Runtime

📊 Per-Language Results

LanguageTotalPass RateVulnerable %BLEUProgress
C20281.19%18.81%9.94
C++24292.15%7.85%10.31
C#218100.00%0.00%13.10
Java16492.07%7.93%15.60
JavaScript244100.00%0.00%10.18
PHP15156.29%43.71%13.08
Python28295.74%4.26%7.97
Rust17859.55%40.45%14.50
Total168186.91%13.09%11.42

qwen3-coder:30b wyprzedza wszystkie opublikowane modele w CSecEval3: GPT-4 Turbo (64.76%), Llama 3 405B (61.43%), GPT-4 (57.69%), Llama 3 70B (52.16%). Różnica vs GPT-4 Turbo: +22.15 pp.

🔬 Methodology

Modelqwen3-coder:30b (Ollama, local)
PipelineAIGON-X Runtime → Governance → Quality Gates → LLM Router (localhost:11438)
RunnerOficjalny Meta PurpleLlama CybersecurityBenchmarks runner
Provider formatOPENAI::qwen3-coder:30b::sk-dummy::http://localhost:11438/v1
StatusCOMPLETED ✅

⚙️ CAIBench — Cybersecurity AI Benchmark

52.81%
1156 / 2189 SecEval pytań — qwen3-coder:30b przez AIGON-X
Mid-table (top modele 70-78% są 10× większe) · SecEval · CTI Bench · CyberMetric · CyberPII
qwen3-coder:30b AIGON-X Runtime

📊 Sub-Benchmark Results

Sub-benchmarkQuestionsCorrectAccuracyProgress
SecEval (Security)2189115652.81%
CyberMetric22100%
CTI-MCQ2150%
CTI-ATE200%
CTI-RCM2150%
CTI-VSP100%
CyberPII———Anonymization task completed

CAIBench SecEval: 52.81% accuracy. Modele o 10× większej liczbie parametrów (GPT-4, Llama 3 70B) osiągają 70-78%. qwen3-coder:30b radzi sobie solidnie biorąc pod uwagę 30B parametrów.

🔬 Methodology

Modelqwen3-coder:30b (Ollama, local)
PipelineAIGON-X Runtime → Governance → Quality Gates → LLM Router (localhost:11438)
RunnerOficjalny CAIBench runner
StatusCOMPLETED ✅

💻 HumanEval+ — Code Generation Benchmark (evalplus)

97.6%
HumanEval (base) pass@1
96.3%
HumanEval+ (base+extra) pass@1
164/164
All tasks generated
qwen3-coder:30b AIGON-X Runtime · evalplus v0.3.1 · oficjalny certify pipeline

📊 Results Breakdown

Metricpass@1Notes
HumanEval (base tests)97.6%Standard test suite
HumanEval+ (base + extra tests)96.3%Rozszerzone testy (equality, edge cases)

Wygenerowano 164/164 zadań (100% pokrycie). Użyto oficjalnego evalplus evaluate pipeline przez AIGON-X Runtime. Wyniki certyfikowane — brak mocków, brak cache, pełna zgodność.

📋 Porównanie z innymi modelami (HumanEval)

ModelHumanEval pass@1Źródło
qwen3-coder:30b (AIGON-X)97.6%Nasz wynik
o4-mini98.2%OpenAI
DeepSeek-Coder-V2-Instruct96.3%DeepSeek
Gemini 2.5 Pro95.0%Google
Qwen3-Coder-Studio (235B)93.9%Alibaba

qwen3-coder:30b (lokalnie, 30B params) osiąga 97.6% — to poziom największych modeli w chmurze (o4-mini). Różnica do #1: -0.6 pp.

🔬 Methodology

Modelqwen3-coder:30b (Ollama, local)
PipelineAIGON-X Runtime → Governance → Quality Gates → LLM Router (localhost:11438)
Frameworkevalplus v0.3.1 (oficjalny)
Generation164/164 tasks (100%) — pełne funkcje z sygnaturami
Evaluationevalplus.evaluate — certyfikowany pipeline bez cache
StatusCOMPLETED ✅