What is preference Elo on CompareLLM?

Preference Elo is a crowd vote from LMArena / Arena. People see two hidden answers and pick the one they like more. The model that wins more often gets a higher Elo. That means people preferred it — not that it passed a school test. It is not SWE-bench, not accuracy, and not a number we invent.

How often does CompareLLM update?

Once a day at 06:00 UTC. A Vercel Hobby cron hits /api/cron/ingest, pulls OpenRouter, LiveBench, official SWE-bench JSON, and Arena Elo, then refreshes pages if a number moved. Humans do not retype cells.

Do you run SWE-bench yourselves?

No. V1 is an aggregator plus optional first-party latency pings. Missing cells stay blank. We do not scrape Artificial Analysis and we do not invent a composite intelligence index.

How do new compare pages appear?

Automatically. Two indexable models that share at least three metrics get /compare/{a}-vs-{b}. Thin pairs still render but are noindex so we do not build doorway pages.

Is public login the same as admin?

No. Anyone can register to vote, comment, and save models. Publishing /news, running ingest by hand, promoting preview models, and attaching aliases is an allowlisted admin desk.

Live Ingest|Updated Aug 17, 2026|101 Active Models

Compare AI Models on Verified Benchmarks, Speed & Real Cost

Name: CompareLLM AI model leaderboard
Creator: CompareLLM
License: https://creativecommons.org/licenses/by/4.0/

Independent crowd preference Elo, SWE-bench coding tests, and live $/1M API pricing with zero synthetic bias.

Best Coding Best Cheap

Intelligence vs API Price Tradeoff

Quality vs. Price Pareto Frontier

Top-Left = Best Value

Plotted by intelligence (Elo) vs cost (Out $). The teal line connects models with unbeatable price-to-performance.

Leaderboard TableTable↓

↑ Up: Crowd vote (Elo)← Left: Cheaper

Vertical ↑

Crowd vote (Elo)

Lost more votes ↓People liked it more ↑

Two hidden answers. A person picks the one they like. Higher Elo means more wins — not a school test or a coding exam.

Horizontal →

Output (answer) price / 1 million tokens (USD)

← CheaperMore expensive →

Cost to generate the answer. Provider list prices are USD

Best-Value Frontier Line (9 Models)Models defining the teal efficiency boundary

Brand:

💬 LLMs 🎨 Text-to-Image ModelsText-to-ImageNew

Text-to-Image Benchmark Ranking

Top Text-to-Image Models (Arena Elo & Speed)

Top 4 of 4 models.Top 4 of 4 models. Rankings from blind preference votes, generation latency, and price per 1k images.

Compare every row vs:

All Models Frontier Open Weights 🇨🇳 China

1FLUX.1 [schnell]Open

Black Forest Labs

Img Elo1,142

Latency1.2s

$/1k$15/1k

Prompt %84%

vs GPT Image 2 (High)Specs

2FLUX.1 [dev]Open

Black Forest Labs

Img Elo1,180

Latency4s

$/1k$30/1k

Prompt %87.5%

vs GPT Image 2 (High)Specs

3Wan2.6 Text to Image🇨🇳 CNOpen

Alibaba

Img Elo1,215

Latency3.1s

$/1k$30/1k

Prompt %89.5%

vs GPT Image 2 (High)Specs

4Ideogram 4.0 Open WeightsOpenFrontier

Ideogram

Img Elo1,218

Latency3.8s

$/1k$60/1k

Prompt %91%

vs GPT Image 2 (High)Specs

#	Model	Image Elo	Gen Time (s)	$ / 1k imgs	Prompt %	Text %
1	FLUX.1 [schnell]Open Black Forest Labs	1,142	1.2s	$15/1k	84%	82%
2	FLUX.1 [dev]Open Black Forest Labs	1,180	4s	$30/1k	87.5%	86%
3	Wan2.6 Text to Image🇨🇳 CNOpen Alibaba	1,215	3.1s	$30/1k	89.5%	87%
4	Ideogram 4.0 Open WeightsOpenFrontier Ideogram	1,218	3.8s	$60/1k	91%	95.8%

4 models in this filter · sorted by image elo

All benchmark leaderboards Intent rankings

US vs China AI FrontierGap Closed & Surpassing on Cost

How Close is China's AI to US Flagships?

Quality is virtually tied on independent benchmarks, while open-weight Chinese frontiers offer substantial price-to-performance advantages.

🇺🇸 US LeaderClaude Opus 5

🇨🇳 China RivalGLM-5.3

General IntelligenceChatbot Arena human preference score

🇺🇸 1624 Elo

🇨🇳 1558 Elo

Near Parity (96%)

Software EngineeringSWE-bench real GitHub bug resolution

🇺🇸 79.2%

🇨🇳 76.4%

Near Parity (96%)

API Pricing & CostOutput price per 1 Million tokens

🇺🇸 $50.00

🇨🇳 $1.80

27.8× Cheaper

Model AvailabilityWeights download & self-hosting

🇺🇸 Closed API

🇨🇳 Open Weights

Open Source

Key Takeaway:

GLM-5.3 matches 96% of Claude Opus 5's intelligence at 96% lower cost.

Full Showdown Fact Sheet

Leaders Matchup

Open GPT Image 2 (High) vs Midjourney v7 fact sheetOpen showdown fact sheet|View all comparisons

Multi-Dimensional Capability Radar

Frontier Trio Capability Radar

Comparing top Western standard models with China's leading frontier rival across 6 skill dimensions. Tap any spoke or dot to inspect.

Tap any node to inspect

0–100 %ile

Verified Intelligence & Dispatches

Recent AI News & Benchmark Briefings

Independent analysis on model promotions, price reductions, and verified score movements.

View all dispatches (51)

GLM-5.2 vs GLM 5V Turbo Price and SWE-bench 2026

news

AIEval EditorialAug 16, 2026

GLM-5.2 vs GLM 5V Turbo Price and SWE-bench 2026

GLM-5.2 undercuts GLM 5V Turbo by 74.2% on input at $0.31/1M, with 1,492 Elo and 69.3% SWE-bench in dated 2026 snapshots.

Shipped & Updated

New and Refreshed Models

View all 460 catalog models

Zhipu2026-08-14

GLM-5.3

Z.ai Aug 14 2026 post-train of the GLM-5.2 744B base. Coding-plan live; open weights promised after a two-week safety review (z.ai/blog/glm-5.3).

Alibaba2026-08-14

Qwen3.8 27B

Alibaba open-weight 27B drop dated Aug 14 2026. Dense enough to self-host; not a frontier MoE.

How to read this leaderboard

Plain-English methodology and leaderboard answers

: Preference Elo is a crowd vote from LMArena / Arena. People see two hidden answers and pick the one they like more. The model that wins more often gets a higher Elo. That means people preferred it — not that it passed a school test. It is not SWE-bench, not accuracy, and not a number we invent.

Compare AI Models on Verified Benchmarks, Speed & Real Cost

Quality vs. Price Pareto Frontier

Top Text-to-Image Models (Arena Elo & Speed)

How Close is China's AI to US Flagships?

Leaders Matchup

Frontier Trio Capability Radar

Recent AI News & Benchmark Briefings

GLM-5.2 vs GLM 5V Turbo Price and SWE-bench 2026

New and Refreshed Models

How to read this leaderboard

Head-to-Head Comparison Studio

Claude Opus 5 vs GPT-5.6 Sol

Claude Sonnet 5 vs DeepSeek V4

Qwen3.8 vs Llama 4 Scout

GLM-5.3 vs Gemini 3.7 Flash

Haiku 4.5 vs Llama 4 Scout: cheap Claude versus cheap open Llama

Qwen3.8 27B (Aug 14): dense enough to self-host, not a frontier MoE