Évaluations objectives et inter-modèles de la sécurité et des capacités. Fondé sur des données d'incidents réels et sur l'arbitrage de débats multimodèles.
Verification of structural instruction-following accuracy using IFEval-like constraints.
Adversarial jailbreak protection and prompt-injection defense rating under human-approved simulation.
Needle-in-a-haystack retrieval evaluation for context lengths up to 32k tokens.
Real-world AI incident evaluation grounded in the ALPAR incident registry, combined with adversarial jailbreaks.
TruthfulQA-based factuality analysis paired with curated Turkish-context factual claims.
Evaluation of Turkish language proficiency using translated MMLU-TR and bespoke local regulatory test sets.
Assessment of model decision-making logic against Art. 73 regulatory scenarios based on EU AI Act taxonomy.
Standard mathematical problem-solving capability checked against GSM8K and MATH benchmarks.
| Rank | Modèle | Fournisseur | K10 | K11 | K12 | K5 | K6 | K7 | K8 | K9 | Score composite |
|---|---|---|---|---|---|---|---|---|---|---|---|
| OpenAI (ChatGPT) | 96% | 95% | 80% | 89% | 95% | 88% | 94% |
| 96% |
| 92% |
| GPT-4o | OpenAI (ChatGPT) | 87% | 93% | 94% | 94% | 90% | 88% | 89% | 90% | 91% |
| GPT-4o mini | OpenAI (ChatGPT) | 91% | 88% | 89% | 93% | 94% | 94% | 90% | 88% | 91% |
| #4 | GPT-4.5 | OpenAI (ChatGPT) | 94% | 86% | 80% | 88% | 90% | 90% | 91% | 93% | 89% |
| #5 | GPT-4o | OpenAI (ChatGPT) | 85% | 88% | 80% | 88% | 92% | 88% | 90% | 93% | 88% |
| #6 | Claude 4.5 Sonnet | Anthropic (Claude) | 86% | 88% | 80% | 87% | 89% | 89% | 81% | 86% | 86% |
| #7 | Claude 4.7 Opus | Anthropic (Claude) | 86% | 82% | 80% | 88% | 86% | 89% | 85% | 88% | 86% |
| #8 | Claude 3.5 Sonnet | Anthropic (Claude) | 84% | 89% | 80% | 83% | 88% | 82% | 85% | 88% | 85% |
| #9 | Gemini 2.0 Flash | Google (Gemini) | 88% | 81% | 88% | 85% | 88% | 87% | 81% | 78% | 85% |
| #10 | Gemini 2.5 Pro | Google (Gemini) | 79% | 89% | 85% | 88% | 87% | 87% | 83% | 83% | 85% |
| #11 | Gemini 3.5 Flash | Google (Gemini) | 79% | 86% | 80% | 83% | 86% | 88% | 86% | 88% | 85% |
| #12 | Claude 4.8 Sonnet | Anthropic (Claude) | 84% | 89% | 80% | 83% | 87% | 87% | 80% | 87% | 85% |
| #13 | Gemini 3.5 Flash | Google (Gemini) | 88% | 78% | 80% | 89% | 89% | 82% | 81% | 89% | 85% |
| #14 | Claude 3.5 Sonnet | Anthropic (Claude) | 82% | 86% | 80% | 89% | 87% | 87% | 86% | 84% | 85% |
| #15 | Claude 4 Sonnet | Anthropic (Claude) | 88% | 85% | 80% | 86% | 89% | 82% | 87% | 80% | 85% |
| #16 | Gemini 2.5 Pro | Google (Gemini) | 85% | 89% | 80% | 87% | 86% | 85% | 83% | 82% | 85% |
| #17 | Claude 3.7 Sonnet | Anthropic (Claude) | 80% | 79% | 87% | 89% | 83% | 84% | 85% | 82% | 84% |
| #18 | Gemini 2.5 Flash | Google (Gemini) | 81% | 79% | 81% | 88% | 81% | 88% | 89% | 84% | 84% |
| #19 | Claude 4.6 Sonnet | Anthropic (Claude) | 88% | 80% | 80% | 86% | 85% | 87% | 81% | 82% | 84% |
| #20 | Gemini 3.0 Flash | Google (Gemini) | 88% | 85% | 80% | 82% | 87% | 84% | 89% | 79% | 84% |
| #21 | Claude 4.5 Sonnet | Anthropic (Claude) | 88% | 85% | 80% | 88% | 80% | 85% | 88% | 78% | 84% |
| #22 | Gemini 1.5 Pro | Google (Gemini) | 83% | 86% | 80% | 80% | 87% | 88% | 88% | 79% | 84% |
| #23 | Gemini 1.5 Flash | Google (Gemini) | 86% | 78% | 80% | 89% | 82% | 83% | 89% | 87% | 84% |
| #24 | Gemini 2.5 Flash | Google (Gemini) | 85% | 84% | 80% | 84% | 87% | 88% | 87% | 79% | 84% |
| #25 | Gemini 3.0 Pro | Google (Gemini) | 87% | 85% | 80% | 80% | 81% | 81% | 87% | 88% | 84% |
| #26 | Gemini 3.0 Flash | Google (Gemini) | 88% | 78% | 80% | 89% | 84% | 89% | 82% | 81% | 84% |
| #27 | Claude 4 Sonnet | Anthropic (Claude) | 88% | 80% | 84% | 84% | 81% | 80% | 82% | 88% | 83% |
| #28 | Claude 4.5 Opus | Anthropic (Claude) | 80% | 88% | 80% | 87% | 83% | 82% | 80% | 80% | 83% |
| #29 | Claude 4.6 Sonnet | Anthropic (Claude) | 84% | 81% | 80% | 84% | 80% | 82% | 84% | 87% | 83% |
| #30 | Claude 4.8 Opus | Anthropic (Claude) | 87% | 79% | 80% | 82% | 80% | 84% | 87% | 86% | 83% |
| #31 | Gemini 3.0 Flash | Google (Gemini) | 79% | 84% | 80% | 88% | 89% | 84% | 80% | 83% | 83% |
| #32 | Gemini 3.5 Pro | Google (Gemini) | 83% | 85% | 80% | 83% | 84% | 86% | 81% | 81% | 83% |
| #33 | Claude 4.5 Opus | Anthropic (Claude) | 79% | 83% | 80% | 86% | 80% | 86% | 88% | 79% | 83% |
| #34 | Claude 4.8 Opus | Anthropic (Claude) | 85% | 83% | 80% | 84% | 80% | 82% | 87% | 84% | 83% |
| #35 | Claude 3.5 Opus | Anthropic (Claude) | 87% | 79% | 80% | 84% | 88% | 80% | 88% | 80% | 83% |
| #36 | Claude 3.7 Sonnet | Anthropic (Claude) | 81% | 84% | 80% | 80% | 89% | 84% | 88% | 81% | 83% |
| #37 | Claude 4.6 Sonnet | Anthropic (Claude) | 84% | 84% | 80% | 85% | 87% | 85% | 80% | 82% | 83% |
| #38 | Claude 4.5 Sonnet | Anthropic (Claude) | 78% | 84% | 80% | 80% | 87% | 85% | 81% | 79% | 82% |
| #39 | Claude 4.7 Opus | Anthropic (Claude) | 79% | 81% | 80% | 84% | 80% | 83% | 87% | 78% | 82% |
| #40 | Claude 4.8 Sonnet | Anthropic (Claude) | 78% | 84% | 80% | 81% | 84% | 84% | 85% | 83% | 82% |
| #41 | Claude 4.8 Sonnet | Anthropic (Claude) | 78% | 78% | 80% | 86% | 84% | 84% | 86% | 80% | 82% |
| #42 | Claude Fable 5 | Anthropic (Claude) | 81% | 80% | 80% | 85% | 87% | 84% | 80% | 78% | 82% |
| #43 | Gemini 2.0 Flash | Google (Gemini) | 82% | 78% | 80% | 81% | 83% | 86% | 87% | 79% | 82% |
| #44 | Gemini 3.5 Pro | Google (Gemini) | 78% | 78% | 80% | 85% | 83% | 86% | 80% | 78% | 81% |
| #45 | Devin 2.0 | Cognition AI | - | - | 80% | - | - | - | - | - | 80% |
| #46 | Devin 3.0 | Cognition AI | - | - | 80% | - | - | - | - | - | 80% |
| #47 | Llama 4.5 70B | Meta (Llama) | 76% | 76% | 80% | 76% | 65% | 79% | 76% | 76% | 76% |
| #48 | Llama 3.1 70B | Meta (Llama) | 76% | 79% | 80% | 77% | 65% | 74% | 77% | 76% | 76% |
| #49 | Llama 3 8B | Meta (Llama) | 78% | 68% | 80% | 73% | 69% | 72% | 78% | 78% | 75% |
| #50 | Pixtral 12B | Mistral AI | 74% | 70% | 80% | 71% | 76% | 79% | 74% | 79% | 75% |
| #51 | Mistral Large 2 | Mistral AI | 70% | 79% | 80% | 74% | 66% | 76% | 70% | 76% | 74% |
| #52 | GPT-5 | OpenAI (ChatGPT) | 74% | 72% | 80% | 79% | 78% | 71% | 67% | 67% | 74% |
| #53 | o5 | OpenAI (ChatGPT) | 69% | 73% | 80% | 76% | 76% | 75% | 71% | 74% | 74% |
| #54 | Llama 4.5 405B | Meta (Llama) | 75% | 78% | 80% | 75% | 76% | 69% | 72% | 65% | 74% |
| #55 | Llama 5 Scout | Meta (Llama) | 76% | 68% | 80% | 79% | 74% | 69% | 75% | 73% | 74% |
| #56 | Grok 4 | xAI (Grok) | 73% | 74% | 80% | 69% | 70% | 74% | 78% | 75% | 74% |
| #57 | Gemma 2 9B | Google (Gemini) | 70% | 74% | 80% | 74% | 76% | 67% | 77% | 74% | 74% |
| #58 | Gemma 3 8B | Google (Gemini) | 79% | 79% | 80% | 79% | 68% | 68% | 69% | 67% | 74% |
| #59 | Veo 3.1 | Google (Gemini) | 62% | 76% | 80% | 75% | 74% | 68% | 77% | 76% | 74% |
| #60 | Llama 4 Scout | Meta (Llama) | 75% | 65% | 80% | 79% | 73% | 72% | 69% | 76% | 74% |
| #61 | Grok 4 | xAI (Grok) | 62% | 75% | 80% | 72% | 75% | 79% | 78% | 74% | 74% |
| #62 | DeepSeek-Coder-V2 | DeepSeek | 73% | 71% | 80% | 68% | 74% | 79% | 79% | 66% | 74% |
| #63 | Mistral Large 2 | Mistral AI | 60% | 79% | 80% | 77% | 66% | 78% | 78% | 73% | 74% |
| #64 | Codestral | Mistral AI | 70% | 75% | 80% | 76% | 76% | 70% | 66% | 79% | 74% |
| #65 | Llama 4 Maverick | Meta (Llama) | 67% | 71% | 80% | 79% | 69% | 75% | 77% | 65% | 73% |
| #66 | Grok 3 mini | xAI (Grok) | 66% | 67% | 80% | 76% | 74% | 75% | 75% | 67% | 73% |
| #67 | Mistral Small 3.2 | Mistral AI | 78% | 74% | 80% | 76% | 73% | 66% | 72% | 64% | 73% |
| #68 | Llama 4.5 405B | Meta (Llama) | 65% | 78% | 80% | 75% | 65% | 73% | 77% | 73% | 73% |
| #69 | Grok 4.5 | xAI (Grok) | 77% | 75% | 80% | 73% | 75% | 74% | 66% | 60% | 73% |
| #70 | Qwen 3.0 | Qwen | 61% | 79% | 80% | 72% | 68% | 68% | 75% | 79% | 73% |
| #71 | o3 | OpenAI (ChatGPT) | 64% | 66% | 80% | 79% | 73% | 74% | 70% | 79% | 73% |
| #72 | Llama 3.1 405B | Meta (Llama) | 76% | 74% | 80% | 71% | 79% | 71% | 68% | 63% | 73% |
| #73 | Llama 3.2 3B | Meta (Llama) | 68% | 79% | 80% | 76% | 71% | 69% | 74% | 70% | 73% |
| #74 | Qwen 2.5 Coder 32B | Alibaba (Qwen) | 70% | 72% | 80% | 74% | 77% | 72% | 69% | 70% | 73% |
| #75 | Mistral Large 3 | Mistral AI | 64% | 65% | 80% | 78% | 75% | 71% | 76% | 74% | 73% |
| #76 | Grok 3 | xAI (Grok) | 78% | 67% | 80% | 65% | 78% | 71% | 68% | 67% | 72% |
| #77 | GPT-5 | OpenAI (ChatGPT) | 65% | 76% | 80% | 70% | 73% | 70% | 67% | 77% | 72% |
| #78 | DeepSeek-V4 | DeepSeek | 68% | 69% | 80% | 74% | 69% | 79% | 75% | 61% | 72% |
| #79 | Qwen 2.5 | Qwen | 70% | 79% | 80% | 68% | 67% | 72% | 75% | 66% | 72% |
| #80 | Qwen 2.5 | Qwen | 68% | 64% | 80% | 68% | 75% | 74% | 78% | 69% | 72% |
| #81 | Qwen 3.0 | Qwen | 75% | 62% | 80% | 71% | 79% | 70% | 70% | 66% | 72% |
| #82 | GPT-5.1 Codex Max | OpenAI (ChatGPT) | 69% | 70% | 80% | 78% | 73% | 67% | 76% | 64% | 72% |
| #83 | Sora 2 | OpenAI (ChatGPT) | 79% | 65% | 80% | 66% | 75% | 69% | 67% | 71% | 72% |
| #84 | o4-mini | OpenAI (ChatGPT) | 75% | 63% | 80% | 72% | 77% | 74% | 66% | 70% | 72% |
| #85 | Gemma 4 31B | Google (Gemini) | 75% | 62% | 80% | 70% | 66% | 73% | 74% | 72% | 72% |
| #86 | Llama 3 70B | Meta (Llama) | 79% | 72% | 80% | 79% | 71% | 68% | 67% | 61% | 72% |
| #87 | Llama 4 Maverick | Meta (Llama) | 62% | 79% | 80% | 77% | 72% | 78% | 69% | 60% | 72% |
| #88 | Grok 4.5 | xAI (Grok) | 67% | 61% | 80% | 78% | 78% | 71% | 72% | 71% | 72% |
| #89 | DeepSeek-R1 | DeepSeek | 63% | 74% | 80% | 74% | 78% | 66% | 74% | 63% | 72% |
| #90 | Qwen 2.5 72B | Alibaba (Qwen) | 67% | 78% | 80% | 71% | 71% | 69% | 72% | 65% | 72% |
| #91 | Qwen 3.0 | Alibaba (Qwen) | 68% | 65% | 80% | 75% | 69% | 76% | 76% | 65% | 72% |
| #92 | Codestral 2.0 | Mistral AI | 69% | 61% | 80% | 75% | 77% | 68% | 69% | 77% | 72% |
| #93 | o3 | OpenAI (ChatGPT) | 67% | 70% | 65% | 74% | 66% | 74% | 75% | 73% | 71% |
| #94 | Llama 3.3 70B | Meta (Llama) | 78% | 78% | 63% | 67% | 68% | 67% | 75% | 71% | 71% |
| #95 | GPT-5 mini | OpenAI (ChatGPT) | 73% | 77% | 80% | 66% | 68% | 69% | 75% | 61% | 71% |
| #96 | Llama 4.5 70B | Meta (Llama) | 67% | 61% | 80% | 75% | 76% | 65% | 69% | 77% | 71% |
| #97 | Grok 4 | xAI (Grok) | 76% | 76% | 80% | 70% | 67% | 71% | 65% | 66% | 71% |
| #98 | GPT-5 mini | OpenAI (ChatGPT) | 73% | 61% | 80% | 71% | 79% | 74% | 69% | 63% | 71% |
| #99 | o4 | OpenAI (ChatGPT) | 70% | 63% | 80% | 71% | 65% | 78% | 73% | 68% | 71% |
| #100 | Grok 4.5 | xAI (Grok) | 65% | 73% | 80% | 72% | 69% | 73% | 66% | 71% | 71% |
| #101 | DeepSeek-V3 | DeepSeek | 70% | 71% | 80% | 66% | 65% | 69% | 74% | 73% | 71% |
| #102 | DeepSeek-V4 | DeepSeek | 63% | 79% | 80% | 68% | 66% | 66% | 68% | 76% | 71% |
| #103 | GPT-5.2 | OpenAI (ChatGPT) | 69% | 76% | 80% | 67% | 66% | 66% | 66% | 78% | 71% |
| #104 | o3-mini | OpenAI (ChatGPT) | 61% | 79% | 80% | 67% | 69% | 65% | 68% | 79% | 71% |
| #105 | Gemma 2 27B | Google (Gemini) | 65% | 75% | 80% | 73% | 69% | 66% | 67% | 69% | 71% |
| #106 | Llama 4.5 70B | Meta (Llama) | 67% | 61% | 80% | 79% | 66% | 72% | 69% | 73% | 71% |
| #107 | Grok 2 | xAI (Grok) | 76% | 60% | 80% | 66% | 69% | 76% | 68% | 70% | 71% |
| #108 | Grok 3 | xAI (Grok) | 66% | 76% | 80% | 71% | 73% | 70% | 66% | 63% | 71% |
| #109 | DeepSeek-V2.5 | DeepSeek | 63% | 62% | 80% | 76% | 76% | 67% | 65% | 75% | 71% |
| #110 | DeepSeek-V3.2 | DeepSeek | 75% | 62% | 80% | 74% | 73% | 65% | 68% | 72% | 71% |
| #111 | Qwen 2 7B | Alibaba (Qwen) | 75% | 64% | 80% | 72% | 67% | 79% | 68% | 63% | 71% |
| #112 | Qwen 2 72B | Alibaba (Qwen) | 60% | 71% | 80% | 75% | 69% | 69% | 74% | 69% | 71% |
| #113 | Mistral NeMo | Mistral AI | 68% | - | - | 70% | 69% | 72% | 70% | 77% | 71% |
| #114 | o4-mini | OpenAI (ChatGPT) | 71% | 77% | 64% | 69% | 74% | 70% | 69% | 67% | 70% |
| #115 | o4 | OpenAI (ChatGPT) | 65% | 66% | 80% | 73% | 67% | 67% | 72% | 68% | 70% |
| #116 | Llama 5 Scout | Meta (Llama) | 61% | 68% | 80% | 66% | 74% | 69% | 66% | 79% | 70% |
| #117 | DeepSeek-V3 | DeepSeek | 61% | 65% | 80% | 77% | 67% | 75% | 70% | 64% | 70% |
| #118 | Llama 3.3 70B | Meta (Llama) | 65% | 71% | 80% | 76% | 68% | 70% | 66% | 60% | 70% |
| #119 | Llama 4.5 405B | Meta (Llama) | 60% | 60% | 80% | 78% | 66% | 68% | 76% | 73% | 70% |
| #120 | DeepSeek-V4 Flash | DeepSeek | 68% | 62% | 80% | 77% | 72% | 65% | 76% | 62% | 70% |
| #121 | Grok 3 | xAI (Grok) | 77% | 70% | 65% | 66% | 70% | 72% | 70% | 60% | 69% |
| #122 | Llama 4 Scout | Meta (Llama) | 71% | 69% | 60% | 74% | 66% | 77% | 72% | 65% | 69% |
| #123 | o5 | OpenAI (ChatGPT) | 78% | 61% | 80% | 70% | 66% | 65% | 67% | 67% | 69% |
| #124 | GPT-5.2 Instant | OpenAI (ChatGPT) | 63% | 61% | 80% | 68% | 71% | 69% | 73% | 70% | 69% |
| #125 | Gemma 4 9B | Google (Gemini) | 70% | 68% | 80% | 73% | 65% | 67% | 70% | 62% | 69% |
| #126 | Llama 3.1 8B | Meta (Llama) | 62% | 62% | 80% | 69% | 79% | 66% | 73% | 63% | 69% |
| #127 | Grok 3 mini | xAI (Grok) | 61% | 68% | 80% | 73% | 72% | 65% | 73% | 60% | 69% |