AI model benchmarks

Compare AI models through Arena preference charts, capability benchmarks, cost trade-offs, rankings and original evaluation evidence.

Reports, evaluations & safety observations

Report statistics, measured evaluations and reported incidents retain their own meaning.

AI Incident Database · 1,801

Stanford AI Index · 2

AILuminate · 32

Hugging Face Evaluations · 2,078

LiveCodeBench · 35

τ-bench · 85

Terminal-Bench · 61

ClockQA · exact match

AI Index 2026 · 2. Technical Performance · exact match · percent

Saxena et al., 2025

  • Qwen2-VL-7B0% 0%
  • Llama 3.2-Vision3.23% 3.23%
  • MiniCPM-V-2.63.23% 3.23%
  • GPT-o14.84% 4.84%
  • Claude-3.5-sonnet6.45% 6.45%
  • GPT-4o8.06% 8.06%
  • Gemini-2.022.58% 22.58%
Official report and methodology ↗

Technical performance relative to human baselines

AI Index 2026 · 2. Technical Performance · performance relative to human baseline · relative to human baseline

AI Index, 2026

20122025Human baseline = 1AIMEGPQA DiamondImageNet Top-5MATHMMLUMMMUOSWorldSQuAD 2.0SWE-bench VerifiedSuperGLUEVQA
  • AIME
  • GPQA Diamond
  • ImageNet Top-5
  • MATH
  • MMLU
  • MMMU
  • OSWorld
  • SQuAD 2.0
  • SWE-bench Verified
  • SuperGLUE
  • VQA
Official report and methodology ↗

Artificial Analysis

artificialAnalysis · Artificial Analysis · Agentic Index · index

  1. Anthropic: Claude Fable 5.1anthropic57.2
  2. Anthropic: Claude Opus 5anthropic56.5
  3. Qwen: Qwen3.8 Max (0902)qwen56
  4. Meta: Muse Spark 1.3meta-llama55.5
  5. Z.ai: GLM 5.3z-ai53.1
  6. SpaceXAI: Grok 4.6x-ai52.5
  7. OpenAI: GPT-6 Astraopenai51
  8. Z.ai: GLM 5.3 Flashz-ai50.9
  9. Anthropic: Claude Fable 5anthropic50.7
  10. Qwen: Qwen3.8 2.4T A95Bqwen50.1
  11. DeepSeek: DeepSeek V4 Flash Vision Expdeepseek47.5
  12. OpenAI: GPT-5.6 Solopenai47.4

Artificial Analysis

artificialAnalysis · Artificial Analysis · Coding Index · index

  1. Anthropic: Claude Fable 5.1anthropic80.7
  2. OpenAI: GPT-5.6 Solopenai78.3
  3. Anthropic: Claude Opus 5anthropic78
  4. OpenAI: GPT-6 Astraopenai76.9
  5. OpenAI: GPT-5.6 Terraopenai76.7
  6. Anthropic: Claude Fable 5anthropic76.5
  7. Google: Gemini 3.8 Flashgoogle76.3
  8. Qwen: Qwen3.8 Max (0902)qwen76.2
  9. Google: Gemini 3.7 Flashgoogle76.1
  10. SpaceXAI: Grok 4.6x-ai75.9
  11. Meta: Muse Spark 1.3meta-llama75.8
  12. OpenAI: GPT-5.5openai74.9

Artificial Analysis

artificialAnalysis · Artificial Analysis · Intelligence Index · index

  1. Anthropic: Claude Fable 5.2anthropic64.6
  2. Google: Gemini 4 Carbongoogle60
  3. Anthropic: Claude Opus 5.5anthropic59.1
  4. OpenAI: GPT-6 Astraopenai56.9
  5. OpenAI: GPT-6.1 Solopenai54.6
  6. Anthropic: Claude Sonnet 5.5anthropic54.5
  7. OpenAI: GPT-6.1 Terraopenai51.3
  8. Google: Gemini 4 Argongoogle50.4
  9. Anthropic: Claude Fable 5.1anthropic50
  10. Anthropic: Claude Fable 5anthropic49.6
  11. OpenAI: GPT-5.6 Solopenai47
  12. MoonshotAI: Kimi K3.1 Promoonshotai46.5

BFCL Irrelevance Detection

bfcl · BFCL · Irrelevance detection · percent

  1. nvidia/llama-3.1-nemotron-ultra-253b-v1nvidia100%
  2. mistralai/Ministral-8B-Instruct-2410mistralai100%
  3. claude-haiku-4-5@20251001anthropic95.29%
  4. Anthropic: Claude Sonnet 4.5anthropic95.03%
  5. Google: Gemini 2.5 Flashgoogle93.67%
  6. Google: Gemini 2.5 Flash Litegoogle93.33%
  7. DeepSeek: DeepSeek V3.2 Expdeepseek93.18%
  8. Google: Gemini 2.5 Flash Litegoogle92.5%
  9. Google: Gemini 2.5 Flashgoogle91.09%
  10. OpenAI: GPT-5 Miniopenai91.01%
  11. claude-opus-4-5@20251101anthropic90.75%
  12. OpenAI: GPT-5 Nanoopenai89.1%

BFCL Live

bfcl · BFCL · Live accuracy · percent

  1. gemini-3-pro-previewgoogle83.12%
  2. Qwen: Qwen3 32Bqwen82.01%
  3. Qwen: Qwen3 32Bqwen82.01%
  4. gemini-3-pro-previewgoogle81.72%
  5. Anthropic: Claude Sonnet 4.5anthropic81.13%
  6. Z.ai: GLM 4.6z-ai80.9%
  7. Qwen: Qwen3 14Bqwen80.01%
  8. claude-opus-4-5@20251101anthropic79.79%
  9. Nanbeige/Nanbeige4-3B-Thinking-2511nanbeige79.42%
  10. Qwen: Qwen3 14Bqwen79.35%
  11. OpenAI: GPT-4.1openai78.9%
  12. claude-haiku-4-5@20251001anthropic78.68%

BFCL Memory

bfcl · BFCL · Memory accuracy · percent

  1. claude-opus-4-5@20251101anthropic73.76%
  2. Anthropic: Claude Sonnet 4.5anthropic64.95%
  3. gemini-3-pro-previewgoogle61.72%
  4. Z.ai: GLM 4.6z-ai55.7%
  5. gemini-3-pro-previewgoogle54.84%
  6. claude-haiku-4-5@20251001anthropic54.41%
  7. DeepSeek: DeepSeek V3.2 Expdeepseek54.19%
  8. Grok 4.1 Fast Reasoningspacexai53.98%
  9. OpenAI: o3openai51.83%
  10. OpenAI: o3openai47.31%
  11. OpenAI gpt-5.2 · azure-openai v2025-12-11openai45.81%
  12. OpenAI: GPT-5 Miniopenai44.3%

BFCL Multi Turn

bfcl · BFCL · Multi-turn accuracy · percent

  1. Salesforce/xLAM-2-32b-fc-rsalesforce69.5%
  2. claude-opus-4-5@20251101anthropic68.38%
  3. Z.ai: GLM 4.6z-ai68%
  4. gemini-3-pro-previewgoogle63.12%
  5. OpenAI: o3openai62.25%
  6. Anthropic: Claude Sonnet 4.5anthropic61.37%
  7. gemini-3-pro-previewgoogle60.75%
  8. Grok 4.1 Fast Reasoningspacexai58.87%
  9. Salesforce/xLAM-2-3b-fc-rsalesforce58.38%
  10. claude-haiku-4-5@20251001anthropic53.62%
  11. Nanbeige/Nanbeige4-3B-Thinking-2511nanbeige51.12%
  12. Qwen: Qwen3 32Bqwen47.87%

BFCL Non-Live AST

bfcl · BFCL · Non-live AST accuracy · percent

  1. gemini-3-pro-previewgoogle90.65%
  2. Qwen: Qwen3 32Bqwen90.27%
  3. claude-opus-4-5@20251101anthropic89.65%
  4. Salesforce/xLAM-2-32b-fc-rsalesforce89.6%
  5. Qwen: Qwen3 14Bqwen89.46%
  6. Meta: Llama 4 Scoutmeta-llama89.38%
  7. Qwen: Qwen3 30B A3B Instruct 2507qwen88.92%
  8. Qwen: Qwen3 32Bqwen88.77%
  9. OpenAI: GPT-4.1openai88.69%
  10. Anthropic: Claude Sonnet 4.5anthropic88.65%
  11. claude-opus-4-5@20251101anthropic88.58%
  12. Grok 4.1 Fast Reasoningspacexai88.27%

BFCL Overall

bfcl · BFCL · Overall accuracy · percent

  1. claude-opus-4-5@20251101anthropic77.47%
  2. Anthropic: Claude Sonnet 4.5anthropic73.24%
  3. gemini-3-pro-previewgoogle72.51%
  4. Z.ai: GLM 4.6z-ai72.38%
  5. Grok 4.1 Fast Reasoningspacexai69.57%
  6. claude-haiku-4-5@20251001anthropic68.7%
  7. gemini-3-pro-previewgoogle68.14%
  8. OpenAI: o3openai63.05%
  9. Grok 4.1 Fast Non-Reasoningspacexai58.29%
  10. DeepSeek: DeepSeek V3.2 Expdeepseek56.73%
  11. Google: Gemini 2.5 Flashgoogle56.24%
  12. OpenAI gpt-5.2 · azure-openai v2025-12-11openai55.87%

BFCL Relevance Detection

bfcl · BFCL · Relevance detection · percent

  1. tiiuae/Falcon3-7B-Instructtiiuae100%
  2. OpenAI: GPT-4.1openai100%
  3. Meta: Llama 4 Scoutmeta-llama100%
  4. speakleash/Bielik-11B-v2.3-Instructspeakleash93.75%
  5. DeepSeek: DeepSeek V3.2 Expdeepseek93.75%
  6. tiiuae/Falcon3-10B-Instructtiiuae93.75%
  7. OpenAI: GPT-4.1 Nanoopenai93.75%
  8. OpenAI: GPT-5 Miniopenai93.75%
  9. OpenAI: GPT-5 Nanoopenai93.75%
  10. OpenAI: o3openai93.75%
  11. Qwen: Qwen3 30B A3B Instruct 2507qwen93.75%
  12. Qwen: Qwen3 32Bqwen93.75%

BFCL Web Search

bfcl · BFCL · Web-search accuracy · percent

  1. claude-opus-4-5@20251101anthropic84.5%
  2. claude-haiku-4-5@20251001anthropic83.5%
  3. Grok 4.1 Fast Reasoningspacexai82.5%
  4. OpenAI: GPT-5 Miniopenai82%
  5. Anthropic: Claude Sonnet 4.5anthropic81%
  6. gemini-3-pro-previewgoogle80%
  7. Z.ai: GLM 4.6z-ai77.5%
  8. OpenAI: o3openai77%
  9. OpenAI gpt-5.2 · azure-openai v2025-12-11openai75.5%
  10. OpenAI: o4 Mini (batch)openai75.5%
  11. Grok 4.1 Fast Non-Reasoningspacexai75%
  12. OpenAI: GPT-5 Nanoopenai72.5%

Throughput (tokens/s) bs=50

bigCode · BigCode Models Leaderboard · Throughput (tokens/s) bs=50 · tokensPerSecond · Higher is better

  1. stabilityai/stablecode-completion-alpha-3bstabilityai718
  2. codellama/CodeLlama-7b-hfcodellama693
  3. codellama/CodeLlama-7b-Instruct-hfcodellama693
  4. codellama/CodeLlama-7b-Python-hfcodellama693
  5. codellama/CodeLlama-13b-hfcodellama0
  6. codellama/CodeLlama-13b-Instruct-hfcodellama0
  7. codellama/CodeLlama-13b-Python-hfcodellama0
  8. codellama/CodeLlama-34b-hfcodellama0
  9. codellama/CodeLlama-34b-Instruct-hfcodellama0
  10. codellama/CodeLlama-34b-Python-hfcodellama0
  11. Phind/Phind-CodeLlama-34B-Python-v1phind0
  12. WizardLM/WizardCoder-Python-34B-V1.0wizardlm0

Models

Loading
EVIDENCE / COMPARISON

Benchmarks

Explore every benchmark

Loading

These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

Install TensorTrace