The AI leaderboard with receipts.

Independent scores, prices, speed and real-world usage for 499 models from 154 labs. Every number links to its source.

Top 5 by Intelligence Index
  1. 1Claude Fable 5.2$26.54 to run
  2. 2Gemini 4 Carbon$5.48 to run
  3. 3Claude Opus 5.5$10.04 to run
  4. 4GPT-6 Astra$6.24 to run
  5. 5GPT-6.1 Sol$1.27 to run
Updated 15 Nov 2026. Index and cost to run it from Artificial Analysis.
499models tracked
154labs and providers
92benchmarks audited
2,773sourced scores
174speech models

The whole field. One chart.

Intelligence against price, speed and size for every measured model. Hover any dot to see where it came from.

Model Intelligence & Value Frontier

Artificial Analysis Intelligence Index against blended API price or output speed. Only models with both measurements are plotted.

Best value near the top: Gemini 4 Carbon scores 60.0 (within 15% of the leader) at $4.00 per 1M blended tokens.
Full Leaderboard Table →

Frontier models, right now.

The four highest Intelligence Index scores, side by side.

AnthropicNov 2026
Frontier

Claude Fable 5.2

64.6Intelligence Index
Context
1M
Speed
102 tok/s
Size
Undisclosed
$8.00/$40per 1M tokens
AnthropicSep 2026
Frontier

Claude Opus 5.5

59.1Intelligence Index
Context
1M
Speed
114 tok/s
Size
Dense / Undisclosed
$4.00/$20per 1M tokens
OpenAISep 2026
Frontier

GPT-6 Astra

56.9Intelligence Index
Context
1M
Speed
40 tok/s
Size
Multi-Trillion MoE (Astra Tier)
$10/$50per 1M tokens
Best proprietary64.6Claude Fable 5.2
18.1 ptsgap between the best closed and the best open-weight model. 36 of 57 measured models are open-weight.See the gap over time
Best open-weight46.5Kimi K3.1 Pro

What changed this month.

Releases from the last 30 days and the biggest usage movers on OpenRouter this week.

New releases

Right column: Intelligence Index, where measured.

Biggest usage gains this week

Week-over-week change in tokens among the top 15 models on OpenRouter.

Every leaderboard, at a glance.

The top five on each board. Every board has its own page with the full list.

Benchmarks wear out. We track it.

How many of the 92 tracked benchmarks still separate frontier models. Saturated suites no longer tell models apart.

Open the trust matrix

The labs, and the agents they power.

Each lab ranked by its best measured model, and the coding agents measured end to end.

Top labs

  1. 1AnthropicClaude Fable 5.2 · 21 models, 0 open64.6
  2. 2Google DeepMindGemini 4 Carbon · 51 models, 11 open60.0
  3. 3OpenAIGPT-6 Astra · 65 models, 2 open56.9
  4. 4Meta AIMuse Wonder Full · 25 models, 21 open50.1
  5. 5Moonshot AIKimi K3.1 Pro · 10 models, 4 open46.5
  6. 6xAIGrok 4.7 · 10 models, 0 open35.5
  7. 7TencentHy4 · 8 models, 7 open34.6
  8. 8Alibaba Cloud / QwenQwen3.8 Max (0902) · 56 models, 52 open33.7

Coding agents, measured

AgentModel usedIndexCost / taskTime / task
Claude Code AnthropicClaude Fable 5.2 (Xhigh, Default Fallback)58.5$15.4753 min
Google Antigravity GoogleGemini 4 Carbon (High)55.5$7.1334 min
Muse Code MetaMuse Wonder Full (Max)54.5—34 min
Codex OpenAIGPT-6 Astra (Max)53.1$7.6326 min
Kimi Code CLI KimiKimi K3.1 Pro48.6$2.9442 min

Artificial Analysis Coding Agent Index: FrontierCode-AA 2.0, Terminal-Bench 4.1 and SWE-Atlas-QnA.

Every number has a source.

Scores and release dates come from Artificial Analysis, prices and usage from OpenRouter, benchmark facts from the original papers. Nothing is estimated.