Quality & Reasoning

Overall Arena EloPrimary

LMSYS crowd human preference ranking

Coding Elo & SWE-benchCode

Real GitHub issue software solve rate

LiveBench Reasoning

Contamination-free automated tests

PhD-level science & domain knowledge

Speed & Token Cost

Throughput (tok/s)Speed

Output generation token rate

Time to First Token (TTFT)

Response latency for voice & chat loops

Output Price ($/1M tokens)

Cost per million generated tokens

Live Pareto Frontier Scatter

Quality vs Cost efficiency boundary

Model Classes

Frontier ModelsProprietary

Opus 4.5, GPT-5, Gemini 2.0 Pro

Open-Weight CatalogApache/MIT

Llama 3.3, DeepSeek, Qwen 2.5

🇨🇳 Chinese LLMsCN

DeepSeek V3, Qwen, GLM-5, MiniMax

Browse All 40+ Models

Top Providers

Claude Opus 4.5, Sonnet 4.5, Haiku

GPT-5, GPT-4.5, GPT-4o, o3

DeepSeek V3, R1 Reasoning

Gemini 2.0 Pro, Flash, Thinking

Popular Head-to-Head ShowdownsView all 48+ pairs →

🇨🇳 DeepSeek V3 vs 🇺🇸 GPT-5

East vs West frontier battle

Sonnet 4.5 vs 🇨🇳 DeepSeek V3

Everyday developer favorite

🇨🇳 Qwen 2.5 vs 🇺🇸 Llama 3.3

Open-weights value clash

Claude Opus 4.5 vs GPT-5

Flagship proprietary duel

Open Interactive Comparison Matrix

Best of & Stacks

Best LLM Lists (2026)

Best Coding LLM

SWE-bench verified repository tests

Sub-$1/1M token value powerhouses

Fastest Low-Latency LLM

Sub-200ms TTFT for voice & live chat

Claude vs GPT Benchmark

Anthropic vs OpenAI head-to-head

Stack Engine Presets

Cheap Coding Agents

Budget repo bots with high SWE-bench

Lowest-Latency Chat

Fast interactive support loops

Open-Weight Reasoning

Self-hostable reasoning power

Explore All 11 Presets

Research & News

Intelligence & Telemetry

News & Benchmark BriefingsDispatches

Verified model promotions & price shifts

Hourly Benchmark ChangelogLive

Dated ingest audit trail with exact diffs

Evaluation Methodology

Standardized scoring formulas & harnesses

What is Arena Elo?

Understanding blind pairwise human ratings

…

Precision Benchmarks

Programmatic, dated AI model benchmarks, head-to-head comparisons, and Stack Engine presets.

Daily ingest · 06:00 UTC

Analytics & Benchmarks

AI Model Leaderboard
Head-to-Head Compare Hub
Models Directory
Stack Engine Presets
Frontier Models
Open Weights Catalog

Guides & Intent Lists

Best LLM Lists (2026)
Best Coding LLM
Best Cheap LLM
Fastest Low-Latency LLM
Claude vs GPT Benchmark
What is Elo?
Methodology Guides
News & Dispatches

Transparency & API

Evaluation Methodology
Benchmark Changelog
Public JSON API
llms.txt Specification
Privacy Policy
Sign In / Account

© 2026 CompareLLM. Public benchmark data aggregated from Arena Elo, LiveBench, SWE-bench & OpenRouter.

Every score has a dated snapshot.

News Desk

Editorial Intelligence & Release Desk

AI Model News

34 dispatches published

Deep benchmark briefings, price reductions, and architecture analysis. Every dispatch links verified snapshot data and head-to-head showdowns.

All Launches Prices Versus Analysis

Claude Fable 5: the $10/$50 long-horizon Anthropic SKU

Jun 15, 2026·CompareLLM Intelligence Desk

Claude Fable 5: the $10/$50 long-horizon Anthropic SKU

Fable 5 is not Sonnet with extra adjectives. It is the expensive long-horizon row. Read SWE-bench and output price together.

PrevPage 2 of 4Next

DeepSeek V4 Pro: high reasoning density per dollar

launch

Jun 12, 2026·CompareLLM Intelligence Desk

DeepSeek V4 Pro: high reasoning density per dollar

2026-06 open-weight-adjacent flagship. The usual cheap-vs-Claude question starts here.

deepseek-v4-pro

Read briefing

Kimi K3: Moonshot’s long-context open-weight competitive row

Jun 8, 2026·CompareLLM Intelligence Desk

Kimi K3: Moonshot’s long-context open-weight competitive row

Jun 2026. The Kimi vs GPT hub remaps to the top Moonshot Elo vs top OpenAI Elo.

Claude Opus 4.6: the May follow-on that still shows up in old bookmarks

May 10, 2026·CompareLLM Intelligence Desk

Claude Opus 4.6: the May follow-on that still shows up in old bookmarks

Slightly behind 4.5 on one official SWE-bench sweep; stronger long-horizon traces. We keep the row for old vs URLs.

claude-opus-4-6

Qwen 3 Max: Alibaba’s flagship for math and multilingual code

Apr 18, 2026·CompareLLM Intelligence Desk

Qwen 3 Max: Alibaba’s flagship for math and multilingual code

Closed flagship. Strong math. Compare to Sol and Opus on GPQA and SWE-bench, not on brand heat.

Llama 4 Scout: long-context open sibling of Maverick

Apr 8, 2026·CompareLLM Intelligence Desk

Llama 4 Scout: long-context open sibling of Maverick

Huge window, cheap hosted lists. The Haiku alternative that is actually open-weight.

llama-4-scoutclaude-haiku-4-5

Llama 4 Maverick: Meta’s natively multimodal open-weight flagship

Apr 8, 2026·CompareLLM Intelligence Desk

Llama 4 Maverick: Meta’s natively multimodal open-weight flagship

Apr 2026. The Llama vs Claude hub uses the top Meta Elo row — usually Maverick.

llama-4-maverick

$Gemini 3 Flash: near-top official SWE-bench at a fraction of Opus$ launch

Mar 16, 2026·CompareLLM Intelligence Desk

Gemini 3 Flash: near-top official SWE-bench at a fraction of Opus

March 2026 speed SKU. Still a useful cheap-coding compare even after 3.6/3.7 Flash.

Gemini 3 Pro’s multi-million context is still a 2026 RAG shortlist

Mar 15, 2026·CompareLLM Intelligence Desk

Gemini 3 Pro’s multi-million context is still a 2026 RAG shortlist

March 2026 Google frontier multimodal with a 2M window. 3.6 Pro is newer; 3 Pro stays as the original 3.x Pro row.

Claude Opus 4.5 and the Feb 2026 SWE-bench row we still cite

Feb 20, 2026·CompareLLM Intelligence Desk

Claude Opus 4.5 and the Feb 2026 SWE-bench row we still cite

The Feb refresh made 4.5 a coding reference. Newer Opus SKUs exist; this row stays as the dated harness point.

claude-opus-4-5