AISOTA
EN中文
· State of the Art

Daily AI Decision Intelligence

Multi-source rankings blended into one score. See which model is truly State of the Art — today.

Data as of 2026-09-22Updated Sep 23, 05:00 AM

Overall Intelligence

Multi-source blend of general intelligence benchmarks into a single AISOTA score.

Sources:Artificial AnalysisLiveBench
#ModelArtificial AnalysisLiveBenchAISOTA Score
1SOTAclaude-fable-5-max-effort—96.896.8
22Claude Fable 5.1 (max with fallback)2 sources90.098.494.2
33Claude Opus 5.5 (max with fallback)2 sources95.091.993.5
4GPT-6 Astra (max)2 sources85.095.290.1
5gpt-5.5-xhigh—87.187.1
6Muse Spark 1.3 (max)2 sources75.093.584.3
7gemini-3.7-flash-high—83.983.9
8Claude Opus 5 (max)2 sources80.085.582.8
9smaug-agentic—80.680.6
10GPT-5.6 Sol (max)2 sources65.090.377.7

Coding Ability

Multi-source blend of coding benchmarks into a single AISOTA score.

Sources:Artificial Analysis AgenticLiveBench Coding
#ModelArtificial Analysis AgenticLiveBench CodingAISOTA Score
1SOTAdeepseek-v4.1-flash-max—98.498.4
22claude-opus-5-5-xhigh-effort—96.896.8
33claude-fable-5-1-max-effort—95.295.2
4smaug-agentic—93.593.5
5claude-fable-5-max-effort—91.991.9
6claude-opus-5-max-effort—90.390.3
7muse-spark-1.3-xhigh—88.788.7
8kimi-k3—87.187.1
9glm-5.3—85.585.5
10qwen3.8-max—83.983.9

Value for Money

Intelligence per dollar, based on median input + output price.

Sources:AA IntelligenceOpenRouter Price
#ModelAA IntelligenceOpenRouter PriceAISOTA Score
1SOTAGLM-5.3-Flash2 sources25.090.157.5
22MiMo-V2.6-Pro2 sources55.059.157.0
33Claude Opus 5.5 (max with fallback)2 sources95.013.654.3
4Muse Spark 1.3 (max)2 sources75.025.750.4
5Claude Fable 5.1 (max with fallback)2 sources90.04.647.3
6GPT-6 Sol (max)2 sources70.023.546.8
7GPT-6 Luna (max)2 sources5.088.546.8
8DeepSeek V4.1 Flash (max)2 sources15.078.046.5
9GPT-6 Astra (max)2 sources85.05.345.1
10Claude Opus 5 (max)2 sources80.09.945.0

Model Usage

API request volume per model over the last 7 days, sourced from OpenRouter public rankings.

Sources:OpenRouter Usage
#ModelOpenRouter UsageAISOTA Score
1SOTAdeepseek-v4-flash934.22M99.8
22glm-5-3-flash422.53M99.6
33gpt-5-6-luna305.55M99.3
4deepseek-v4-1-flash281.82M99.1
5gemini-2-5-flash-lite232.65M98.9
6jev-1-13227.7M98.7
7text-embedding-3-small150.41M98.5
8gemini-3-1-flash-lite143.21M98.3
9gpt-oss-120b142.95M98.0
10gemini-2-5-flash128.01M97.8

AI Tools Leaderboard

Ranked lists of the best AI tools across the most popular categories, with editorial notes on each tool.

AI Compute & API Pricing

Live AI compute prices: model API inference costs, cloud GPU benchmark pricing, and best value models per quality point. Updated daily from public sources.

API Inference Pricing

3204 entries

  1. 1SOTAdeepinfra/meta-llama/Llama-3.2-3B-Instruct$0.02/M
  2. 2lambda_ai/llama3.2-3b-instruct$0.02/M
  3. 3nscale/Qwen/Qwen2.5-Coder-3B-Instruct$0.02/M
  4. 4nscale/Qwen/Qwen2.5-Coder-7B-Instruct$0.02/M
  5. 5nebius/Qwen/Qwen2.5-Coder-7B$0.02/M

Chinese AI Models Pricing

1003 entries

  1. 1SOTAnscale/Qwen/Qwen2.5-Coder-3B-Instruct$0.02/M
  2. 2nscale/Qwen/Qwen2.5-Coder-7B-Instruct$0.02/M
  3. 3nebius/Qwen/Qwen2.5-Coder-7B$0.02/M
  4. 4together_ai/Qwen/Qwen2-1.5B-Instruct$0.02/M
  5. 5nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-8B$0.03/M

Cloud GPU Benchmark Pricing

18 entries

  1. 1SOTAStandard_NV4as_v4$0.054/h
  2. 2Standard_NC36ds_xl_RTXPRO6000BSE_v6$0.333/h
  3. 3Standard_NV24s_v3$0.584/h
  4. 4Standard_NV16as_v4$1.901/h
  5. 5Standard_NC320lds_xl_RTXPRO6000BSE_v6$2.643/h

How AISOTA scoring works

Each model is scored against the sources on that board. Within each source, models are ranked and converted to a 0–100 percentile. A model's AISOTA score is the weighted average of its percentile ranks across sources. SOTA marks today's #1.

Data is collected automatically from public APIs each day. Missing cells mean the source does not list that model.