Skip to content
llmwaves

AI trends

What actually changed

Two things move at once: the ceiling keeps rising, and the floor price of any given capability keeps collapsing. The second one changes more decisions than the first.

Frontier record holder

58.1

Qwen3.8 Max · Aug 2026

Index points gained

+51.4

since May 2023

Price of Index 40+

89× cheaper

since it first appeared

Frontier records set

16

distinct record holders

The ceiling

Frontier intelligence over time

Best Intelligence Index score available on any given date. Flat stretches are the interesting part — that is how long a record held.

Frontier Intelligence Index

Single series · a step means a new record holder

01020304050202420252026Intelligence Index

Hover the plot for the model holding the record at that date.

View as table
DateModelIndex
May 2023GPT-4 (OpenAI)6.7
Apr 2024GPT-4 Turbo (OpenAI)7.0
May 2024GPT-4o (OpenAI)8.4
Dec 2024o1 (OpenAI)15.2
Apr 2025o4 Mini (OpenAI)16.7
Apr 2025o3 (OpenAI)20.2
May 2025Claude Opus 4 (Anthropic)26.0
Dec 2025GPT-5.2 (OpenAI)30.4
Feb 2026Qwen3 Max Thinking (Qwen)32.5
Mar 2026GPT-5.4 (OpenAI)39.0
Apr 2026Claude Opus 4.7 (Anthropic)40.7
Apr 2026Qwen3.6 Max Preview (Qwen)41.1
May 2026Claude Opus 4.8 (Anthropic)41.8
Jun 2026Claude Fable 5 (Anthropic)49.6
Jul 2026Claude Opus 5 (Anthropic)50.8
Aug 2026Qwen3.8 Max (Qwen)58.1

The floor

What a capability costs over time

Cheapest blended price at which each intelligence tier was available. Each line steps down when somebody undercuts the previous best.

Cheapest price per intelligence tier

Blended USD per 1M tokens on a log scale · lower is better

  • Index 30+
  • Index 40+
$0.1$1$10Jan 2026AprJulOctBlended USD / 1M tokensIndex 30+Index 40+

A tier only appears once some model reached it. Log scale, because the drops are order-of-magnitude. Where one model is the cheapest option for two tiers at once, those lines coincide — hover to read every tier at a date.

View as table
TierDateModelBlended / 1M
Index 30+Dec 2025GPT-5.2$4.81
Index 30+Feb 2026Qwen3 Max Thinking$1.56
Index 30+Mar 2026KAT-Coder-Pro V2$0.525
Index 30+Apr 2026V4 Flash 0423$0.114
Index 30+Oct 2026Ling 3.1 Flash$0.112
Index 40+Apr 2026Claude Opus 4.7$10
Index 40+Apr 2026Qwen3.6 Max Preview$2.31
Index 40+Jun 2026Nex-N2-Pro$0.438
Index 40+Aug 2026GLM 5.3 Flash$0.237
Index 40+Oct 2026Ling 3.1 Flash$0.112

The race

Six creators, best-so-far

Each line is one creator's own record over time. Convergence at the right edge is the story of the last year.

Best Intelligence Index per creator

Six series · each line steps when that creator beats its own record

  • Anthropic
  • OpenAI
  • Google
  • Qwen
  • DeepSeek
  • Meta
01020304050202420252026Intelligence IndexQwenAnthropicOpenAIGoogleDeepSeekMeta

Only creators with more than one benchmarked release appear. Lines are direct-labelled at the right edge.

View as table
CreatorDateModelIndex
AnthropicMar 2024Claude 3 Haiku3.5
AnthropicMay 2025Claude Sonnet 416.6
AnthropicMay 2025Claude Opus 426.0
AnthropicFeb 2026Claude Opus 4.626.4
AnthropicApr 2026Claude Opus 4.740.7
AnthropicMay 2026Claude Opus 4.841.8
AnthropicJun 2026Claude Fable 549.6
AnthropicJul 2026Claude Opus 550.8
AnthropicSep 2026Claude Fable 5.153.4
AnthropicSep 2026Claude Opus 5.557.6
OpenAIMay 2023GPT-46.7
OpenAIApr 2024GPT-4 Turbo7.0
OpenAIMay 2024GPT-4o8.4
OpenAIDec 2024o115.2
OpenAIApr 2025o4 Mini16.7
OpenAIApr 2025o320.2
OpenAIJun 2025o3 Pro21.9
OpenAIAug 2025GPT-523.0
OpenAINov 2025GPT-5.1-Codex23.7
OpenAINov 2025GPT-5.124.7
OpenAIDec 2025GPT-5.230.4
OpenAIFeb 2026GPT-5.3-Codex32.5
OpenAIMar 2026GPT-5.439.0
OpenAIJul 2026GPT-5.6 Sol47.0
OpenAISep 2026GPT-6 Astra52.7
GoogleJun 2025Gemini 2.5 Pro16.1
GoogleFeb 2026Gemini 3.1 Pro Preview29.7
GoogleMay 2026Gemini 3.5 Flash32.6
GoogleJul 2026Gemini 3.6 Flash34.0
GoogleAug 2026Gemini 3.7 Flash39.1
GoogleSep 2026Gemini 3.8 Flash40.9
GoogleOct 2026Gemini 4 Argon52.6
QwenSep 2024Qwen2.5 72B Instruct7.7
QwenJul 2025Qwen3 Coder 30B A3B Instruct9.6
QwenSep 2025Qwen3 Max24.5
QwenFeb 2026Qwen3 Max Thinking32.5
QwenApr 2026Qwen3.6 Max Preview41.1
QwenAug 2026Qwen3.8 Max58.1
DeepSeekJan 2025R113.1
DeepSeekSep 2025V3.1 Terminus13.9
DeepSeekDec 2025V3.216.0
DeepSeekApr 2026V4 Flash 042334.3
DeepSeekApr 2026V4 Pro 042336.0
DeepSeekSep 2026V4.1 Flash39.5
MetaApr 2025Llama 4 Scout6.5
MetaApr 2025Llama 4 Maverick9.3

Cadence

Releases and standings

How many models shipped per quarter, and where each creator's best model currently sits.

Models released per quarter

Count of models entering the index · last twelve quarters

  • 2024-Q1
    4
  • 2024-Q2
    5
  • 2024-Q3
    17
  • 2024-Q4
    14
  • 2025-Q1
    24
  • 2025-Q2
    28
  • 2025-Q3
    46
  • 2025-Q4
    45
  • 2026-Q1
    47
  • 2026-Q2
    57
  • 2026-Q3
    98
  • 2026-Q4
    12

Counts the models tracked here, not every model released anywhere.

View as table
QuarterModels
2024-Q14
2024-Q25
2024-Q317
2024-Q414
2025-Q124
2025-Q228
2025-Q346
2025-Q445
2026-Q147
2026-Q257
2026-Q398
2026-Q412

Best model per creator

Intelligence Index of each creator's strongest release

One entry per creator — this is a standings table, not a model ranking.

View as table
CreatorBest modelIndex
QwenQwen3.8 Max58.1
AnthropicClaude Opus 5.557.6
OpenAIGPT-6 Astra52.7
GoogleGemini 4 Argon52.6
MetaMuse Wonder52.4
xAIGrok 4.746.4
XiaomiMiMo-V2.6-Pro46.3
Z.aiGLM 5.344.8
StepFunStep 5 Preview43.7
Moonshot AIKimi K343.6
Nex AGINex-N2-Pro41.7
InclusionAILing 3.1 Flash41.1