AI Benchmark Leaderboard

Compare AI model scores, coding ability and performance in one place.

# Model Org Intelligence ↓ CodingMathMMLU ProGPQALiveCodeBenchAIME 2025MATH 500SciCodeIFBenchHLE
76 Ling 3.1 Flash 1019 InclusionAI 29.6 ——————————
77 GPT-6.1 Terra (Low) OpenAI 29.4 ——————————
78 MiniMax-M3.1-Flash MiniMax 29.1 ——————————
79 DeepSeek V4.1 Flash 1017 (High) DeepSeek 28.6 ——————————
80 Qwen3.8-Flash-Next Alibaba 28.6 ——————————
81 Step 5 Preview StepFun 28.5 ——————————
82 GPT-6 Luna (Max) OpenAI 27.9 ——————————
83 GLM-5.3-Flash Z AI 27.8 ——————————
84 Gemma 4 26B A4B (Non-reasoning) Google 27.1 29.1 —— 71.4 ——— 37.3 45.4 10.7
85 Gemini 4 Hydrogen (Low) Google 26.9 ——————————
86 Agnes 2.5 Pro Alpha (based on Qwen3.5 397B A17B) Sapiens AI 26.8 58.8 —— 87.6 ——— 42.9 — 33.6
87 MiMo-V2.6-Flash Xiaomi 26.7 ——————————
88 Mistral Large 4 (High) Mistral 26.1 ——————————
89 K2 Horizon MoVA 36B A4B MBZUAI Institute of Foundation Models 25.7 ——— 82.2 ——— 40.0 — 23.4
90 Claude Sonnet 5.5 (Low, Default Fallback) Anthropic 25.6 ——————————
91 Mistral Large 4 Preview Mistral 25.3 ——————————
92 Muse Spark Open (Xhigh) Meta 24.9 ——————————
93 Claude Haiku 5.5 (Medium) Anthropic 24.6 ——————————
94 Ling 3.0 Tiny InclusionAI 24.3 26.5 —— 73.4 ——— 24.2 — 9.3
95 Agnes 2.5 Pro Theta Sapiens AI 24.0 ——————————
96 GPT-5.6 Luna (Max) OpenAI 23.8 ——————————
97 GPT-6 Luna (Xhigh) OpenAI 23.5 ——————————
98 DeepSeek V4 Pro 0813 (Max) DeepSeek 23.1 ——————————
99 GLM-5.3 (Low) Z AI 23.1 ——————————
100 Mistral Large 4 (Medium) Mistral 23.1 ——————————

AI Benchmarks FAQ

Which AI model is best for my task?

Compare the scores most relevant to your task, then consider speed, pricing and practical requirements. A high score in one benchmark does not make a model the best choice for every use.

What does a missing benchmark score mean?

A dash means that a score is not available for that model in this table. It should not be treated as a score of zero.

How are the top models ranked?

By the intelligence index, highest first. The Top models cards show the eight highest-scoring current models, one per model family, and models with the same score share a place.

Where do the scores come from?

From independent third-party model evaluations, which Franklin AI syncs automatically each day. Each model page links to About these benchmarks, which explains them in more detail.

What does a model page show?

Its scores, its price per million input and output tokens, its measured speed and comparisons with other models, where that data is available.