The independent LLM leaderboard

How the composite score is calculated

Each benchmark is normalised onto a 0–100 scale (Arena Elo is rescaled from a 1000–1500 band; percent benchmarks pass through unchanged). The composite is a weighted average across the available benchmarks for each model. Models with fewer than three published benchmarks are listed without a composite score so that single-benchmark coding specialists cannot displace well-rounded frontier models.

Scores are compiled from provider technical reports, public papers, and Chatbot Arena snapshots. Submit corrections via the GitHub issue tracker. Browse all models A–Z →

Popular guides & head-to-heads

• Best LLM for coding (2026)guideSWE-Bench & HumanEval ranked, with API price and context window for every tier.
• Best LLM for RAG (2026)guideLong-context recall, citation faithfulness, and price-per-1M ranked head-to-head.
• Best LLM for agents (2026)guideTool-call reliability, SWE-Bench, and per-task cost for production agentic workloads.
• Best open-source LLM (2026)guideDeepSeek R1, Llama 3.3, Qwen, and Phi-4 ranked by benchmarks and licence.
• Best cheap LLM API (2026)guidePrice-per-quality rankings, hidden fees, and cost calculators.
• Best free LLM API (2026)guideReal free tiers ranked: Gemini, DeepSeek, Groq, OpenRouter — with rate limits and the catch.
• ChatGPT vs Claude vs GeminivsThree-way frontier head-to-head: benchmarks, pricing, and verdict by use case.
• GPT-5 vs Claude Opus 4.1vsFrontier head-to-head: benchmarks, pricing, and verdict by use case.
• Claude vs GeminivsAnthropic's precision vs Google's scale and 2M context.
• GPT-5 vs Gemini 2.5 ProvsOpenAI's flagship vs Google's 2M-context multimodal model.
• DeepSeek R1 vs GPT-5vsThe best open-weights reasoning model against the closed frontier.
• Claude Sonnet 4 vs GPT-4.1vsMid-tier production workhorses compared.
• Llama 3.3 70B vs Qwen 2.5 72BvsThe two strongest open-weights generalists at 70B scale.

Popular head-to-head comparisons

Pick any two models — composite score, raw benchmark numbers, API pricing, and a one-click route to try both behind the same key.

• GPT-5 vs Claude Opus 4.1frontierThe two highest-scoring closed-frontier models — by composite, code, and price.
• GPT-5 vs Gemini 2.5 ProfrontierOpenAI flagship vs Google's 2M-context multimodal frontier.
• GPT-5 vs DeepSeek R1closed vs openThe best closed model against the best open-weights reasoning model.
• GPT-5 vs Grok 4frontierOpenAI vs xAI on benchmarks, context window, and price.
• Claude Opus 4.1 vs Gemini 2.5 ProfrontierAnthropic's top model against Google's 2M-context flagship.
• o3 vs DeepSeek R1reasoningOpenAI's reasoning specialist vs the open-weights challenger.
• Claude 3.7 Sonnet vs GPT-5frontierMid-frontier Sonnet against the new top dog.
• GPT-5 mini vs Gemini 2.5 Flashfast / cheapProduction-tier fast models head-to-head on price-per-quality.
• GPT-4o mini vs Gemini 2.0 Flashfast / cheapThe two cheapest mainstream APIs at high volume.
• DeepSeek V3 vs GPT-4o minicheapOpen-weights value vs OpenAI's smallest production model.
• Llama 3.1 405B vs DeepSeek V3openThe two heavyweight open-weights models compared.
• Claude 3.5 Sonnet vs GPT-4ogeneralThe 2024-era workhorses still in heavy production use today.