LLM Pricing
PricingLeaderboardsToolsProvidersReleasesGuides

© 2026 LLM Pricing

About
·Contact
·Privacy
·RSS

Leaderboards

Data from Artificial Analysis · Intelligence Index v? · 0 models with data

Value frontier →

Artificial Analysis Intelligence Index. Score uses the “representative tier” (highest-scoring reasoning tier per model).

Leaders of every board

The top 8 on each board, with the metric that ranks them. Open any model for its full pricing across providers.

Intelligence

  • Claude Fable 5.1AA 65.7
  • Claude Opus 5AA 63.1
  • Claude Fable 5AA 62.1
  • GPT-5.6 SolAA 60.9
  • Grok 4.6AA 60.9
  • Kimi K3AA 59.7
  • GLM-5.3AA 59.5
  • Qwen3.8 MaxAA 58.1

Coding

  • Claude Fable 5.181.6
  • Claude Opus 578
  • GPT-5.6 Sol77.4
  • Grok 4.676.8
  • GPT-5.6 Terra76.7
  • Claude Fable 576.5
  • Kimi K376.2
  • Gemini 3.7 Flash76.1

Agentic

  • Claude Fable 5.161.3
  • Claude Opus 559.2
  • GLM-5.359.1
  • Grok 4.658.7
  • Qwen3.8 Max58.4
  • GLM-5.3-Flash58.2
  • GPT-5.6 Sol57.8
  • Qwen3.8 2.4T A95B57.1

Value

  • DeepSeek V4 Flash 0731576
  • GLM-5.3-Flash484
  • DeepSeek V4 Flash453
  • Ling 3.0 Flash420
  • GPT OSS 120B371
  • Hy3 preview300
  • Qwen3.8 Flash Next294
  • MiMo-V2.5217

Cost per task

  • Gemma 3 27B IT$0
  • Command A$0
  • North Mini Code$0
  • LFM2.5 2.6B$0
  • ling-3.0-tiny$0
  • MiMo-V2.5$0.0104
  • Llama 4 Scout$0.0106
  • Granite 4.2 8B$0.0158

Speed

  • Celeris 11520 tok/s
  • Mercury 2684 tok/s
  • Gemini 2.5 Flash-Lite374 tok/s
  • Ling 3.0 Flash374 tok/s
  • Gemini 3.5 Flash Lite347 tok/s
  • Gemini 3.1 Flash Lite319 tok/s
  • Trinity Large Thinking309 tok/s
  • Nemotron 3.5 Lightning309 tok/s

Usage

  • DeepSeek V4 Flash50949269.57M
  • GPT-5.6 Luna39785504.32M
  • Hy4 preview30052912.71M
  • MiMo-V2.528514920.56M
  • Hy327718824.14M
  • GLM-5.3-Flash25954362M
  • DeepSeek V4 Flash 042321862674.04M
  • Nemotron 3 Ultra (free)17407323.68M

Reading the leaderboards

Each board ranks the same catalog on a different axis: raw intelligence, coding and agentic scores from Artificial Analysis, output speed, cost per task, real 30-day usage, and value. The value board divides intelligence by blended price, with an intelligence floor so a cheap but weak model cannot top it on price alone.

The price basis is switchable because the lowest channel is almost always a third-party gateway, while the official list price reflects the first-party rate. Cost per task can reorder the ranking sharply: a reasoning model with a low unit price still runs up a high bill when it emits many thinking tokens per answer.