AI Leaderboard — Compare 300+ Top AI Models by Intelligence, Speed & Price

On the LLM Stats Leaderboard, Claude Mythos Preview currently leads on GPQA Diamond (94.6% gpqa), the most discriminating reasoning benchmark at the frontier. This AI leaderboard ranks models by the LLM Stats Score, which aggregates GPQA, SWE-Bench Verified, coding-arena performance and pricing into one comparable AI ranking. Rankings refresh continuously as new benchmark results land.

"Best" depends on what you're optimizing for. For frontier reasoning, Claude Mythos Preview leads on GPQA. For coding agents, Claude Opus 4.6 is the strongest in head-to-head coding-arena play. For low cost at frontier quality, Qwen3.7 Max is the cheapest in the top 10 at $1.25 /M tok. The leaders summary above the table names the current winner per axis.

The leading LLMs in 2026 are Claude Mythos Preview, Claude Opus 4.6, and the frontier models from OpenAI (GPT-5 family), Anthropic (Claude Opus and Sonnet), Google (Gemini 3 Pro), xAI (Grok 4), DeepSeek (V3 / R1) and Z.AI (GLM-5). Open-weights leaders include Llama, Qwen and DeepSeek. The full ranking is in the leaderboard table above.

Qwen3.7 Max is the cheapest model in the top 10 by GPQA Diamond, at $1.25 /M tok input. The Cheapest filter on the leaderboard restricts to verified, currently-available frontier models — pricing is pulled from each provider's public price list and cross-checked against billing samples through the LLM Stats proxy.

Grok 4 Fast currently exposes the largest practical context window at 2.0M tokens tokens. Larger context lets you keep more documents, conversation history and tool traces in a single request. For long-document workloads, also consult the per-model "effective context" notes on each model detail page — providers vary in how well they actually use the upper end of their advertised windows.

Mercury 2 currently has the highest output throughput at 925 tok/s. Output speed is measured by routing standardized prompts through each provider's API and averaging tokens-per-second over a 7-day rolling window. Fast inference matters most for streaming chat UIs and agentic loops; for batched async workloads, blended price per 1M tokens is usually the better axis.

Kimi K2.6 currently leads among open-weights LLMs (90.5% gpqa). The open-weights ecosystem is dominated by Llama, Qwen, DeepSeek, Mistral, GLM and Gemma. The dedicated Open LLM Leaderboard filters this catalog to models with publicly released weights so you can self-host or fine-tune.

The LLM Stats Score is a composite that blends verified benchmark results (GPQA Diamond, SWE-Bench Verified, coding-arena), live performance metrics (output throughput, time-to-first-token) and per-token pricing into one comparable number. Pricing and metadata revalidate hourly; live performance updates on a 7-day rolling average. For the full weighting and refresh cadence see the LLM Stats Score methodology. 309 canonical models are tracked across every major lab and inference provider.