AI Agents & Models benchmarks

How performance is measured at the frontier · as of 2026-10-29 · 7 benchmarks

Capability

LMArena Text Arena

Arena Score

Style-controlled, crowd-voted head-to-head preference rating.

↑ higher is betterLMArena

#01
Gemini 4 Carbon (High)
1,528
#02
Gemini 4 Argon (High)
1,520
#03
Claude Opus 5.5 (High)
1,509
#04
Muse Wonder (Max)
1,507
#05
Claude Fable 5.1 (Max)
1,504
#06
Claude Fable 5 (High)
1,504
#07
Claude Opus 4.6 (High)
1,503
#08
Claude Opus 4.7 (High)
1,501
#09
Claude Opus 4.6
1,496
#10
Gemini 3.8 Flash (High)
1,496
#11
Muse Spark 1.3 (Max)
1,495
#12
Claude Opus 4.7
1,494

LiveBench

% score

Objective, contamination-resistant evaluation across reasoning, coding, mathematics, data analysis, language, and instruction following.

↑ higher is betterLiveBench

#01
claude-fable-5.1
88.0%
#02
claude-opus-5.5
86.5%
#03
gpt-6-astra
86.1%
#04
gpt-6.1-sol

SWE-bench Verified

% resolved

Share of real GitHub issues the model resolves end-to-end.

↑ higher is betterSWE-bench

#01
OpenHands + GPT-5
71.8%
#02
Moatless Tools + Claude 4 Sonnet
70.8%
#03
OpenHands + Claude 4 Sonnet
70.4%
#04
SWE-agent + Claude 4 Sonnet

Agents

LMArena Agent Arena

IPS score

Real-session agent quality across task outcomes, steerability, recovery, and tool use.

↑ higher is betterLMArena Agent Arena

#01
Claude Opus 5.5 (Max)
0.209
#02
Gemini 4 Carbon (High)
0.160
#03
Claude Sonnet 5.5 (Max)
0.138
#04

Reasoning

ARC-AGI-2

% accuracy

Verified accuracy on novel abstract-reasoning tasks from ARC Prize's semi-private set.

↑ higher is betterARC Prize

#01
GPT-6 Astra (Max)
95.00%
#02
GPT-6.1 Sol (Max)
94.17%
#03
GPT-6 Astra (XHigh)
93.33%
#04

Economics

Blended cost

$ / 1M tokens

Published price, 3:1 input:output blend — lower is cheaper.

↓ lower is betterLiteLLM (open pricing)

#01
GPT-5.6 Luna
$0.45
#02
Kimi K2.6
$1.71
#03
DeepSeek V4 Pro
$1.98
#04
GLM-5.2

Context

Context window

tokens

Maximum input the model accepts in one request.

↑ higher is betterLiteLLM (open pricing)

#01
Gemini 3.1 Pro
1M
#02
Gemini 3.5 Flash
1M
#03
GLM-5.2
1M
#04
Claude Opus 4.8

← AI Agents & Models dashboard