Skip to main content

Open data, updated daily

AI Model Rankings: The Smartest, Cheapest and Newest AI Models

Which AI model should you use? Start here. We bring together open test results, human votes and list prices, and explain each number in plain words.

Data updated

Winners at a glance

One plain question each, answered from today's data. Tap a card for the full ranking.

Smarter and cheaper: where is the best value?

Higher means a higher Epoch index score. Further left means cheaper. The line joins the best-value models: nothing that costs the same or less scores as high.
Epoch index score against price per million tokensScatter chart of 107 models. Left to right is price, from $0.01 to $100 per million tokens, each step ten times more. Bottom to top is score. Best-value line, cheapest first: Qwen2.5-Coder ($0.04, score 113); Gemma 3 4B ($0.05, score 115.9); Qwen3-8B ($0.07, score 136.1); gpt-oss-20b ($0.10, score 137.7); Qwen3-14B ($0.12, score 138.2); GPT-5 nano ($0.14, score 139.2).100120140160180$0.01$0.10$1$10$100Claude Haiku 5.5: $0.20 per 1M tokens, score 159.3Claude Opus 5.5: $8 per 1M tokens, score 168.8Gemma 3 4B: $0.05 per 1M tokens, score 115.9, open weightsGPT-5 nano: $0.14 per 1M tokens, score 139.2GPT-6.1 Sol: $4 per 1M tokens, score 168.4gpt-oss-20b: $0.10 per 1M tokens, score 137.7, open weightsQwen2.5-Coder (7B): $0.04 per 1M tokens, score 113, open weightsQwen3-14B: $0.12 per 1M tokens, score 138.2, open weightsQwen3.5-9B: $0.17 per 1M tokens, score 139.3, open weightsQwen3-8B: $0.07 per 1M tokens, score 136.1, open weightsClaude Opus 5.5Claude Haiku 5.5Qwen3.5-9BQwen3-14BGemma 3 4B↖ Smarter and cheaperCapability scoreUS$ per 1M tokens (each step 10x)
Epoch index score against price per million tokensScatter chart of 107 models. Left to right is price, from $0.01 to $100 per million tokens, each step ten times more. Bottom to top is score. Best-value line, cheapest first: Qwen2.5-Coder ($0.04, score 113); Gemma 3 4B ($0.05, score 115.9); Qwen3-8B ($0.07, score 136.1); gpt-oss-20b ($0.10, score 137.7); Qwen3-14B ($0.12, score 138.2); GPT-5 nano ($0.14, score 139.2).100120140160180$0.01$0.10$1$10$100GPT-6 Astra: $20 per 1M tokens, score 168.8Claude Haiku 5.5: $0.20 per 1M tokens, score 159.3Claude Opus 5.5: $8 per 1M tokens, score 168.8Gemma 3 4B: $0.05 per 1M tokens, score 115.9, open weightsGPT-5 nano: $0.14 per 1M tokens, score 139.2GPT-6.1 Sol: $4 per 1M tokens, score 168.4gpt-oss-20b: $0.10 per 1M tokens, score 137.7, open weightsQwen2.5-Coder (7B): $0.04 per 1M tokens, score 113, open weightsQwen3-14B: $0.12 per 1M tokens, score 138.2, open weightsQwen3.5-9B: $0.17 per 1M tokens, score 139.3, open weightsQwen3-8B: $0.07 per 1M tokens, score 136.1, open weightsClaude Opus 5.5GPT-6 AstraClaude Haiku 5.5Qwen3.5-9BGPT-5 nanoQwen3-14BQwen3-8B↖ Smarter and cheaperCapability scorePrice per million tokens (US$). Each step right costs 10 times more
Epoch index score against price per million tokensScatter chart of 107 models. Left to right is price, from $0.01 to $100 per million tokens, each step ten times more. Bottom to top is score. Best-value line, cheapest first: Qwen2.5-Coder ($0.04, score 113); Gemma 3 4B ($0.05, score 115.9); Qwen3-8B ($0.07, score 136.1); gpt-oss-20b ($0.10, score 137.7); Qwen3-14B ($0.12, score 138.2); GPT-5 nano ($0.14, score 139.2).100120140160180$0.01$0.10$1$10$100Claude Fable 5: $20 per 1M tokens, score 163.4Claude Fable 5.1: $20 per 1M tokens, score 166.5Claude Haiku 4.5: $2 per 1M tokens, score 142.2Claude Opus 4: $30 per 1M tokens, score 142.5Claude Opus 4.1: $30 per 1M tokens, score 144.1Claude Opus 4.5: $10 per 1M tokens, score 150.1Claude Opus 4.6: $10 per 1M tokens, score 155.7Claude Opus 4.7: $10 per 1M tokens, score 156.8Claude Opus 4.8: $10 per 1M tokens, score 159Claude Opus 5: $10 per 1M tokens, score 163.9Claude Sonnet 4: $6 per 1M tokens, score 141.6Claude Sonnet 4.5: $6 per 1M tokens, score 146.7Claude Sonnet 4.6: $6 per 1M tokens, score 152.8Claude Sonnet 5: $4 per 1M tokens, score 156.9Claude Sonnet 5.5: $4 per 1M tokens, score 166.2DeepSeek-V3.2: $0.31 per 1M tokens, score 146.1, open weightsDeepSeek-V3.2-Exp: $0.30 per 1M tokens, score 144.9, open weightsDeepSeek V4.1 Flash: $0.52 per 1M tokens, score 155.9, open weightsDeepSeek-V4-Flash: $0.52 per 1M tokens, score 146.1, open weightsDeepSeek V4 Flash 0731: $0.24 per 1M tokens, score 154.8, open weightsDeepSeek-V4-Pro: $1.98 per 1M tokens, score 149.1, open weightsDeepSeek V4 Pro 0813: $1.98 per 1M tokens, score 155.6, open weightsGemini 2.5 Flash (Sep 2025): $0.85 per 1M tokens, score 142.9Gemini 2.5 Pro (Jun 2025): $3.44 per 1M tokens, score 145.1Gemini 3.1 Flash-Lite: $0.56 per 1M tokens, score 144.3Gemini 3.1 Pro: $4.5 per 1M tokens, score 155.3Gemini 3.5 Flash: $3.38 per 1M tokens, score 154.9Gemini 3.5 Flash-Lite: $0.85 per 1M tokens, score 144.9Gemini 3.6 Flash: $1.5 per 1M tokens, score 154.5Gemini 3.7 Flash: $1.5 per 1M tokens, score 158Gemini 3.8 Flash: $1.5 per 1M tokens, score 157.4Gemini 3 Flash: $1.13 per 1M tokens, score 151.9Gemini 3 Pro: $4.5 per 1M tokens, score 153.1Gemini 4 Argon: $4 per 1M tokens, score 166.5Gemma 4 26B A4B: $0.26 per 1M tokens, score 141.9, open weightsGemma 4 31B IT: $0.21 per 1M tokens, score 143, open weightsGLM-4.7: $1 per 1M tokens, score 143.6, open weightsGLM-5: $1.55 per 1M tokens, score 145.8, open weightsGLM-5.1: $2.15 per 1M tokens, score 149.8, open weightsGLM-5.2: $2.15 per 1M tokens, score 151.9, open weightsGLM-5.3: $2.15 per 1M tokens, score 156, open weightsGLM-5.3-Flash: $0.24 per 1M tokens, score 153.1, open weightsGPT-5: $3.44 per 1M tokens, score 150GPT-5.1: $3.44 per 1M tokens, score 149.8GPT-5.2: $4.81 per 1M tokens, score 153.7GPT-5.2 Pro: $57.8 per 1M tokens, score 156.3GPT-5.3 Codex: $4.81 per 1M tokens, score 157.1GPT-5.4: $5.63 per 1M tokens, score 157.3GPT-5.4 Mini: $1.69 per 1M tokens, score 148.7GPT-5.4 Nano: $0.46 per 1M tokens, score 146GPT-5.4 Pro: $67.5 per 1M tokens, score 159.6GPT-5.5: $11.3 per 1M tokens, score 159.7GPT-5.5 Pro: $67.5 per 1M tokens, score 162.9GPT-5.6 Luna: $0.45 per 1M tokens, score 156.5GPT-5.6 Sol: $8 per 1M tokens, score 162.8GPT-5.6 Terra: $4.5 per 1M tokens, score 160.1GPT-5 mini: $0.69 per 1M tokens, score 145.3GPT-5 Pro: $41.3 per 1M tokens, score 150.6GPT-6 Astra: $20 per 1M tokens, score 168.8GPT-6 Luna: $0.20 per 1M tokens, score 156.8GPT-6 Sol: $4 per 1M tokens, score 163.7Grok 4: $6 per 1M tokens, score 146.2Grok 4.20: $1.56 per 1M tokens, score 152.1Grok 4.5: $3 per 1M tokens, score 154.1Grok 4.6: $3 per 1M tokens, score 156.6Grok 4.7: $3 per 1M tokens, score 154.4Inkling: $1.76 per 1M tokens, score 148.4, open weightsInkling-Small: $0.66 per 1M tokens, score 150.3, open weightsKimi K2.5: $1.2 per 1M tokens, score 148.1, open weightsKimi K2.6: $1.71 per 1M tokens, score 150.9, open weightsKimi K2.7 Code: $1.71 per 1M tokens, score 150.1, open weightsKimi K2 Thinking: $1.08 per 1M tokens, score 145.8, open weightsKimi K3: $6 per 1M tokens, score 157.9, open weightsMiniMax-M2.5: $0.52 per 1M tokens, score 146.6, open weightsMiniMax-M2.7: $0.52 per 1M tokens, score 145.8, open weightsMiniMax-M3: $0.52 per 1M tokens, score 146.9, open weightsMistral Large 4: $1.03 per 1M tokens, score 150Muse Spark 1.1: $2 per 1M tokens, score 154.1Muse Spark 1.2: $2 per 1M tokens, score 155.1Muse Spark 1.3: $2 per 1M tokens, score 157.2Muse Spark Open: $2 per 1M tokens, score 155.1, open weightso1: $26.3 per 1M tokens, score 141.7o3: $3.5 per 1M tokens, score 146.7o3-pro: $35 per 1M tokens, score 147.3o4-mini: $1.93 per 1M tokens, score 145.5Qwen 3.6 35B-A3B: $0.42 per 1M tokens, score 143.7, open weightsQwen 3.6 Max (Preview): $2.85 per 1M tokens, score 149.4Qwen 3.6 Plus: $1.13 per 1M tokens, score 147.3Qwen 3.8 27B: $1.09 per 1M tokens, score 151.7, open weightsQwen 3.8 Max: $3 per 1M tokens, score 157.4Qwen3-235B-A22B-Thinking (Jul 2025): $0.95 per 1M tokens, score 143.8, open weightsQwen3.5-35B-A3B: $0.50 per 1M tokens, score 142.5, open weightsQwen3.5 397B-A17B: $1.35 per 1M tokens, score 146.8, open weightsQwen3.6 27B: $1.08 per 1M tokens, score 146.2, open weightsQwen3.7-Max: $3.75 per 1M tokens, score 154Qwen3.7-Plus: $0.63 per 1M tokens, score 147Qwen3-Max: $3.05 per 1M tokens, score 142.3Claude Haiku 5.5: $0.20 per 1M tokens, score 159.3Claude Opus 5.5: $8 per 1M tokens, score 168.8Gemma 3 4B: $0.05 per 1M tokens, score 115.9, open weightsGPT-5 nano: $0.14 per 1M tokens, score 139.2GPT-6.1 Sol: $4 per 1M tokens, score 168.4gpt-oss-20b: $0.10 per 1M tokens, score 137.7, open weightsQwen2.5-Coder (7B): $0.04 per 1M tokens, score 113, open weightsQwen3-14B: $0.12 per 1M tokens, score 138.2, open weightsQwen3.5-9B: $0.17 per 1M tokens, score 139.3, open weightsQwen3-8B: $0.07 per 1M tokens, score 136.1, open weightsClaude Opus 5.5GPT-6 AstraGPT-6.1 SolClaude Haiku 5.5Qwen3.5-9BGPT-5 nanoQwen3-14BQwen3-8BGemma 3 4B↖ Smarter and cheaperCapability scorePrice per million tokens (US$). Each step right costs 10 times more
  • Best-value line: nothing cheaper is as smart
  • Diamond = open weights
  • OpenAI
  • Anthropic
  • Alibaba (Qwen)
  • Google
  • DeepSeek
  • Z.ai
  • xAI
  • Moonshot
  • Meta
  • Mistral
  • Other makers
Prices are blended: three parts input to one part output. Diamonds are open-weights models. Tap a dot to open the model.

The best-value steps, cheapest first

Each step costs more and scores higher. The list keeps models in the top half of the Epoch index, so the very cheapest, weakest models on the line are left out. See the Best value leaderboard.

The best-value steps, cheapest first: no other model costs the same or less and scores as high
ModelPrice per 1MEpoch index
gpt-oss-20bOpenAI · Open weights$0.10137.7
Qwen3-14BAlibaba · Open weights$0.12138.2
GPT-5 nanoOpenAI$0.14139.2
Qwen3.5-9BAlibaba · Open weights$0.17139.3
Claude Haiku 5.5Anthropic$0.20159.3
GPT-6.1 SolOpenAI$4.00168.4
Claude Opus 5.5Anthropic$8.00168.8

Newest models

The latest releases that already have an Epoch score. A very new model can be missing while the tests catch up.

How fast is the best score rising?

Each dot is a model that beat every model released before it. The orange line follows open-weights models only.
Best Epoch Capabilities Index score by release dateStep chart of the best score over time, from 2023 to 2026. All models: 24 new records, now 168.8 (Claude Opus 5.5). Open-weights models: 25 new records, now 157.9 (Kimi K3).1001201401601802023202420252026LLaMA-65B: 110 (2023-02-24)GPT-4 (Mar 2023): 125.7 (2023-03-14)GPT-4 Turbo (Nov 2023): 126.4 (2024-01-25)Claude 3 Opus: 126.9 (2024-02-29)GPT-4 Turbo (Apr 2024): 127.3 (2024-04-09)GPT-4o (May 2024): 129 (2024-05-13)Claude 3.5 Sonnet: 130 (2024-06-20)o1-mini: 135.7 (2024-09-12)o1: 141.7 (2024-12-17)Gemini 2.5 Pro (Mar 2025): 144 (2025-03-31)o3: 146.7 (2025-04-16)o3-pro: 147.3 (2025-06-10)GPT-5: 150 (2025-08-07)GPT-5 Pro: 150.6 (2025-10-07)Gemini 3 Pro: 153.1 (2025-11-18)GPT-5.2 Pro: 156.3 (2025-12-11)GPT-5.3 Codex: 157.1 (2026-02-05)GPT-5.4 Pro: 159.6 (2026-03-05)GPT-5.5 Pro: 162.9 (2026-04-23)Claude Fable 5: 163.4 (2026-06-09)Claude Opus 5: 163.9 (2026-07-24)Claude Fable 5.1: 166.5 (2026-09-01)GPT-6 Astra: 168.8 (2026-09-03)Claude Opus 5.5: 168.8 (2026-09-22)LLaMA-65B: 110 (2023-02-24)Llama 2-70B: 113.7 (2023-07-18)Stable Beluga 2: 117 (2023-07-20)Yi-34B: 117.3 (2023-11-02)Mixtral 8x7B: 118.4 (2023-12-11)Mixtral 8x22B: 122 (2024-04-17)Llama 3-70B: 123 (2024-04-18)DeepSeek-V2 (MoE-236B, May 2024): 124.8 (2024-05-07)Qwen2-72B: 125.3 (2024-06-07)Llama 3.1-405B: 128.8 (2024-07-23)Qwen2.5-72B: 129 (2024-09-19)Phi-4: 130.5 (2024-12-12)DeepSeek-V3: 132.3 (2024-12-26)DeepSeek-R1: 138.8 (2025-01-20)Qwen3-235B-A22B: 139.2 (2025-04-28)DeepSeek-R1 (May 2025): 141.1 (2025-05-28)Qwen3-235B-A22B-Thinking (Jul 2025): 143.8 (2025-07-25)DeepSeek-V3.2-Exp: 144.9 (2025-09-29)Kimi K2 Thinking: 145.8 (2025-11-06)DeepSeek-V3.2: 146.1 (2025-12-01)Kimi K2.5: 148.1 (2026-01-27)GLM-5.1: 149.8 (2026-04-07)Kimi K2.6: 150.9 (2026-04-20)GLM-5.2: 151.9 (2026-06-16)Kimi K3: 157.9 (2026-07-16)Claude Opus 5.5Claude Fable 5DeepSeek-R1Phi-4Best score so far
Best Epoch Capabilities Index score by release dateStep chart of the best score over time, from 2023 to 2026. All models: 24 new records, now 168.8 (Claude Opus 5.5). Open-weights models: 25 new records, now 157.9 (Kimi K3).1001201401601802023202420252026LLaMA-65B: 110 (2023-02-24)GPT-4 (Mar 2023): 125.7 (2023-03-14)GPT-4 Turbo (Nov 2023): 126.4 (2024-01-25)Claude 3 Opus: 126.9 (2024-02-29)GPT-4 Turbo (Apr 2024): 127.3 (2024-04-09)GPT-4o (May 2024): 129 (2024-05-13)Claude 3.5 Sonnet: 130 (2024-06-20)o1-mini: 135.7 (2024-09-12)o1: 141.7 (2024-12-17)Gemini 2.5 Pro (Mar 2025): 144 (2025-03-31)o3: 146.7 (2025-04-16)o3-pro: 147.3 (2025-06-10)GPT-5: 150 (2025-08-07)GPT-5 Pro: 150.6 (2025-10-07)Gemini 3 Pro: 153.1 (2025-11-18)GPT-5.2 Pro: 156.3 (2025-12-11)GPT-5.3 Codex: 157.1 (2026-02-05)GPT-5.4 Pro: 159.6 (2026-03-05)GPT-5.5 Pro: 162.9 (2026-04-23)Claude Fable 5: 163.4 (2026-06-09)Claude Opus 5: 163.9 (2026-07-24)Claude Fable 5.1: 166.5 (2026-09-01)GPT-6 Astra: 168.8 (2026-09-03)Claude Opus 5.5: 168.8 (2026-09-22)LLaMA-65B: 110 (2023-02-24)Llama 2-70B: 113.7 (2023-07-18)Stable Beluga 2: 117 (2023-07-20)Yi-34B: 117.3 (2023-11-02)Mixtral 8x7B: 118.4 (2023-12-11)Mixtral 8x22B: 122 (2024-04-17)Llama 3-70B: 123 (2024-04-18)DeepSeek-V2 (MoE-236B, May 2024): 124.8 (2024-05-07)Qwen2-72B: 125.3 (2024-06-07)Llama 3.1-405B: 128.8 (2024-07-23)Qwen2.5-72B: 129 (2024-09-19)Phi-4: 130.5 (2024-12-12)DeepSeek-V3: 132.3 (2024-12-26)DeepSeek-R1: 138.8 (2025-01-20)Qwen3-235B-A22B: 139.2 (2025-04-28)DeepSeek-R1 (May 2025): 141.1 (2025-05-28)Qwen3-235B-A22B-Thinking (Jul 2025): 143.8 (2025-07-25)DeepSeek-V3.2-Exp: 144.9 (2025-09-29)Kimi K2 Thinking: 145.8 (2025-11-06)DeepSeek-V3.2: 146.1 (2025-12-01)Kimi K2.5: 148.1 (2026-01-27)GLM-5.1: 149.8 (2026-04-07)Kimi K2.6: 150.9 (2026-04-20)GLM-5.2: 151.9 (2026-06-16)Kimi K3: 157.9 (2026-07-16)Claude Opus 5.5Claude Opus 5GPT-5.3 CodexGLM-5.1DeepSeek-R1DeepSeek-V3Best score so far
Best Epoch Capabilities Index score by release dateStep chart of the best score over time, from 2023 to 2026. All models: 24 new records, now 168.8 (Claude Opus 5.5). Open-weights models: 25 new records, now 157.9 (Kimi K3).1001201401601802023202420252026LLaMA-65B: 110 (2023-02-24)GPT-4 (Mar 2023): 125.7 (2023-03-14)GPT-4 Turbo (Nov 2023): 126.4 (2024-01-25)Claude 3 Opus: 126.9 (2024-02-29)GPT-4 Turbo (Apr 2024): 127.3 (2024-04-09)GPT-4o (May 2024): 129 (2024-05-13)Claude 3.5 Sonnet: 130 (2024-06-20)o1-mini: 135.7 (2024-09-12)o1: 141.7 (2024-12-17)Gemini 2.5 Pro (Mar 2025): 144 (2025-03-31)o3: 146.7 (2025-04-16)o3-pro: 147.3 (2025-06-10)GPT-5: 150 (2025-08-07)GPT-5 Pro: 150.6 (2025-10-07)Gemini 3 Pro: 153.1 (2025-11-18)GPT-5.2 Pro: 156.3 (2025-12-11)GPT-5.3 Codex: 157.1 (2026-02-05)GPT-5.4 Pro: 159.6 (2026-03-05)GPT-5.5 Pro: 162.9 (2026-04-23)Claude Fable 5: 163.4 (2026-06-09)Claude Opus 5: 163.9 (2026-07-24)Claude Fable 5.1: 166.5 (2026-09-01)GPT-6 Astra: 168.8 (2026-09-03)Claude Opus 5.5: 168.8 (2026-09-22)LLaMA-65B: 110 (2023-02-24)Llama 2-70B: 113.7 (2023-07-18)Stable Beluga 2: 117 (2023-07-20)Yi-34B: 117.3 (2023-11-02)Mixtral 8x7B: 118.4 (2023-12-11)Mixtral 8x22B: 122 (2024-04-17)Llama 3-70B: 123 (2024-04-18)DeepSeek-V2 (MoE-236B, May 2024): 124.8 (2024-05-07)Qwen2-72B: 125.3 (2024-06-07)Llama 3.1-405B: 128.8 (2024-07-23)Qwen2.5-72B: 129 (2024-09-19)Phi-4: 130.5 (2024-12-12)DeepSeek-V3: 132.3 (2024-12-26)DeepSeek-R1: 138.8 (2025-01-20)Qwen3-235B-A22B: 139.2 (2025-04-28)DeepSeek-R1 (May 2025): 141.1 (2025-05-28)Qwen3-235B-A22B-Thinking (Jul 2025): 143.8 (2025-07-25)DeepSeek-V3.2-Exp: 144.9 (2025-09-29)Kimi K2 Thinking: 145.8 (2025-11-06)DeepSeek-V3.2: 146.1 (2025-12-01)Kimi K2.5: 148.1 (2026-01-27)GLM-5.1: 149.8 (2026-04-07)Kimi K2.6: 150.9 (2026-04-20)GLM-5.2: 151.9 (2026-06-16)Kimi K3: 157.9 (2026-07-16)Claude Opus 5.5Claude Opus 5GPT-5.5 ProGPT-5.2 Proo3-proKimi K3GLM-5.2DeepSeek-V3.2DeepSeek-R1DeepSeek-V3Best score so far
  • All models: now 168.8 (Claude Opus 5.5)
  • Open-weights models: now 157.9 (Kimi K3)
Newest record: Claude Opus 5.5, 168.8 points.

Consensus top 10: where two views agree

One test can be gamed and one vote can be a mood. This table ranks models by how high they sit on both lists at once. It is our own calculation. See how it works.
Top 10 models by consensus score, which averages the Epoch index percentile and the Arena text rating percentile
Consensus rank and modelConsensus scoreEpoch index rankArena text rankBlended price per 1M
1Claude Opus 5.5Anthropic
99.21 of 2702 of 166$8.00
2Gemini 4 ArgonGoogle
98.45 of 2701 of 166$4.00
3Claude Fable 5.1Anthropic
97.54 of 2705 of 166$20.00
4Claude Fable 5Anthropic
96.29 of 2704 of 166$20.00
5Muse WonderMeta
96.012 of 2702 of 166Not measured
6Claude Opus 5Anthropic
94.67 of 27011 of 166$10.00
7Gemini 3.8 FlashGoogle
92.020 of 2708 of 166$1.50
8GPT-6.1 SolOpenAI
91.73 of 27024 of 166$4.00
9GPT-5.6 SolOpenAI
91.411 of 27018 of 166$8.00
10Gemini 3.7 FlashGoogle
91.118 of 27013 of 166$1.50
The exact formula

Consensus score: we take every model that has both an Epoch Capabilities Index score and an Arena human-vote text rating, and work out where it sits in each list as a percentile from 0 to 100 (100 = best of the group). The consensus score is the average of those two percentiles. A model needs both scores to get one, and the percentiles are only measured against other models that also have both. It is our own calculation, not a score published by Epoch or Arena.

All leaderboards

Different questions need different tests. Pick the one that matches your work.

How to read these numbers

  • Epoch Capabilities Index (ECI)

    A single score from Epoch AI that combines results from many benchmarks to rank how capable models are overall.

    It is the quickest way to compare overall capability. It does not tell you about price, speed or fit for one job.

  • Arena rating

    A score built from people choosing which of two anonymous model answers they prefer, turned into a rating similar to chess Elo.

    It shows what real people prefer, which can differ from what tests measure, such as tone and clarity.

  • Blended price

    A single price per million tokens that mixes input and output at a fixed ratio so models can be compared in one number.

    Use it to compare models quickly, then check the real input and output prices for your own mix of work.

  • Context window

    The context window is the amount of text, counted in tokens, that a model can take into account at one time.

    A bigger window lets you work with long documents or large code bases in one go, but filling it costs more and answers can get less reliable on very long inputs.

Questions people ask

Which AI model is the smartest right now?

Claude Opus 5.5 ranks first on the Epoch Capabilities Index with 168.8 points. GPT-6 Astra is next (168.8), then GPT-6.1 Sol (168.4). The leader's likely range overlaps with 9 of the next 9 models, so the top spots are too close to call. The Epoch Capabilities Index combines many hard benchmarks into one number, so one lucky test cannot decide the order.

Which AI model do people like most?

Gemini 4 Argon is people's favourite in Arena's blind votes, with a rating of 1517. Claude Opus 5.5 is next (1509), then Muse Wonder (1509). The leader's likely range overlaps with 2 of the next 9 models, so the top spots are too close to call. The votes come from Arena, where people compare two anonymous answers and pick the better one.

Which strong AI model is the cheapest?

gpt-oss-20b is the cheapest of the 119 strong models we rank, at $0.10 per million tokens (blended price). Qwen3-14B is next ($0.12), then GPT-5 nano ($0.14). These are list prices for each maker's own API, so confirm on the maker's site before you buy.

What is the best open-weights AI model?

Kimi K3 is the top open-weights model on the Epoch Capabilities Index, with 157.9 points. GLM-5.3 is next (156.0), then DeepSeek V4.1 Flash (155.9). The leader's likely range overlaps with 6 of the next 9 models, so the top spots are too close to call. Open weights means the model can be downloaded. Licences differ, and some limit commercial use, so read the licence before you build on one.

Where do these numbers come from, and how often are they updated?

We combine three open sources: Epoch AI's benchmark data, Arena's human-vote ratings and the LiteLLM price list. We check them every day and rank 279 models from what is published. A model moves only when a source publishes new results, and we never fill a gap with a guess.

Where this data comes from

  • Epoch AI: Benchmark data: Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks (CC BY 4.0). We match models across sources, rank them and show them in charts. Scores are not changed. Epoch AI does not endorse this site.
  • Arena (formerly LMArena): Human-preference ratings: Arena (arena.ai) leaderboard dataset, CC BY 4.0. We keep the latest published ratings, match them to models from other sources and rank them. Ratings are not changed. Arena does not endorse this site.
  • LiteLLM model price list: Prices and context windows: LiteLLM model price list (github.com/BerriAI/litellm, MIT licence). We use the model maker's own price entry when there is one. Prices change often, so always confirm on the maker's site before you buy.

Newest data on this page: 26 Oct 2026. Prices are list prices and change often, so confirm on the maker's site before you buy. Sponsors never change a score or an order. How we calculate this · Report a mistake

We use optional cookies for analytics and referral tracking. Learn more about our privacy policy