Models & benchmarks

Compare AI models without losing the evidence

Follow frontier, open and local models, see what benchmark results actually measure, and connect capability to access, hardware and cost.

Model landscape

Frontier and commercial models

Track the major model families, providers, access routes, context windows, published prices and practical strengths without treating marketing claims as benchmarks.

AI model registryCurrent international and Chinese model families in one evidence-backed view.International modelsOpenAI, Anthropic, Google, xAI and other major providers.Chinese AI modelsChina-based model families, access, prices and capabilities.

text modelsreasoningcodingmultimodal AIlong contextagentsconsumer accessAPI access

Run it yourself

Open, local and self-hosted AI

Understand open-weight licensing, local deployment, memory needs and the hardware trade-offs that determine what can realistically run outside a hosted service.

Meta Muse Glimmer and the local-agent raceWhy deployment fit, privacy, latency and control can matter as much as frontier benchmark leadership.Open modelsOpen-weight model choices, licensing and practical fit.Local modelsModels suited to desktops, workstations and local servers.Local AI hardwareMemory, accelerators, price and comfortable model fit.

open weightslicensesquantizationVRAMRAMGPUsNPUsAI PCs

Measure capability

Benchmarks and evaluations

Compare models using separate evidence for different abilities instead of collapsing every task into one misleading score.

AI benchmarksReasoning, mathematics, coding, agents, long context and document understanding.Why ARC-AGI-3 changes AI benchmarkingHow interactive evaluation measures exploration, goal discovery, planning and action efficiency—and why agent architecture now matters to the score.How to identify an AI model properlyWhy family, version, deployment, tools and reasoning settings matter for reproducible comparisons.Reasoning effort explainedWhy inference effort is an experimental variable rather than a different intelligence score.Research & safetyRed teaming, safety evaluations and reliability testing.

reasoningmathcodingagent taskslong contextdocumentsred teamingevaluation limits

Choose intelligently

Capability, access and value

A useful model comparison includes what it can do, what it costs, how it is accessed, whether it can be deployed locally and how strong the supporting evidence is.

Why AI model routing is becoming a systems layerHow per-request routing can trade quality, cost and latency across model pools—and why current routers still miss the best model surprisingly often.AI pricingPublished API input and output pricing.Pricing historyDated record of verified price changes.How AI works explainersContext windows, RAG, fine-tuning, agents and infrastructure explained.

cost per tokenfree accesssubscriptionslatencyprivacydeployment controlmodel fit