INDEPENDENT MODEL COMPARISONS
Which AI model leads?
Real benchmark results, visualized. Explore tasks, compare models, and understand the scores.
15 charts3 published datasets · Source dates through 2026-10-23
General↗
Gemini 4 Argon (High)
1525.7Arena Rating · Arena
Coding↗
Claude Opus 5.5 (High)
1545Arena Rating · Arena
Reasoning↗
Claude Opus 5.5 (High)
1538.5Arena Rating · Arena
Images↗
gpt-image-2.5-sunburst
1424.6Arena Rating · Arena
EXPLORE THE SCORES
General model preference
Model rankings10 models
Tap a bar to compare up to 3 models.
Blind pairwise human preference, not percent accuracy. · Non-zero x-axis highlights differences.
BROWSE BY WHAT YOU NEED
Explore every benchmark
Find the right test for coding, reasoning, visuals and more. Every card opens an interactive comparison.
Arena2026-10-22
General model preference
#1Gemini 4 Argon (High)1525.7
Arena RatingOpen chart
Arena2026-10-22
Mathematical reasoning
#1Gemini 4 Argon (High)1537.8
Arena RatingOpen chart
Arena2026-10-22
Hard prompts
#1Claude Opus 5.5 (High)1538.5
Arena RatingOpen chart
Arena2026-10-22
Coding preference
#1Claude Opus 5.5 (High)1545
Arena RatingOpen chart
Arena2026-10-22
Web development
#1Claude Opus 5.5 (Max)1806.9
Arena RatingOpen chart
Arena2026-10-22
Chinese conversations
#1Gemini 4 Argon (High)1600.2
Arena RatingOpen chart
Arena2026-10-19
Image understanding
#1Claude Fable 5 (High)1324.9
Arena RatingOpen chart
Arena2026-10-07
Text-to-image
#1gpt-image-2.5-sunburst1424.6
Arena RatingOpen chart
Arena2026-10-20
Image editing
#1GPT-Image-2.5 Sunburst1514.4
Arena RatingOpen chart
Arena2026-10-20
Agent capability · IPS
#1Claude Opus 5.5 (Max)0.2092
IPSOpen chart
Arena2026-10-22
Instruction following
#1Claude Opus 5.5 (High)1538.8
Arena RatingOpen chart
MTEB2026-10-23
Scientific retrieval · SciFact
#1Qwen/Qwen3-Embedding-4B78.33
nDCG@10 × 100Open chart
MTEB2026-10-23
Semantic similarity · STS
#1Qwen/Qwen3-Embedding-4B93.7
Spearman × 100Open chart
EvalPlus2026-10-23
HumanEval+ code tests
#1O1 Mini (Sept 2024)89
pass@1 (%)Open chart
EvalPlus2026-10-23
MBPP+ code tests
#1O1 Preview (Sept 2024)80.2
pass@1 (%)Open chart
OUR DATA
Grounded in published results.
We visualize released results from these three publishers. Each chart preserves its own metric and data date. A rank in one test is not an overall ranking.
Other benchmark references & methodology
These are methodology links. We do not reproduce their charts or claims without appropriate source data and reuse rights.