LLM Reference is a weekly‑updated decision‑support directory that helps engineers and technology leaders choose the right LLM and provider for their specific task.
What is LLM Reference?
LLM Reference tracks 1,894 language models from 142 providers and 249 labs, refreshed every week with new releases, verified price changes, and benchmark updates. It takes your use case (coding, agents, writing, research, image, video, etc.) and surfaces editors’ picks, side‑by‑side model comparisons, and the cheapest serving provider. The output is a shortlist of models with their latest benchmarks, pricing per million tokens, and a Pulse feed showing what changed this week.
Key Features
- Editors’ picks by task — curated leaderboards for 18 use cases (e.g., Coding picks Claude Fable 5; Agents picks Claude Sonnet 5; Image picks BFLUX.2 Dev). Each board shows the top model, its benchmark scores, and runner‑ups.
- Side‑by‑side comparison — Compare any two models on a single page, e.g., GPT‑5.6 Sol vs Grok 4.5, with pricing and benchmark rows.
- Weekly Pulse feed — 71 new models, 96 price cuts, and 17 benchmark refreshes tracked in real time, plus frontier output $/1M tokens.
- Search & filter — Full model directory searchable by name, provider, or capability (Coding, RAG, Agents, Long context, Vision, Classification, JSON/Tool use).
- Route per request — A separate gateway/router directory with 17 verified providers that choose the best model for each API call.
- Live shortlist — A “best overall today” and “freshest update” widget that surfaces DeepSeek V4 Flash (coding default) and GLM‑4‑Flash‑250414 (new this week).
Who is it for?
- Engineering teams that need to pick a coding or agent model with verified benchmark scores (e.g., SWE‑bench Pro, OSWorld‑Verified) and the cheapest provider.
- Technology leaders evaluating the weekly model landscape — the Pulse feed and price‑cut log help them decide when to switch providers.
- Knowledge workers doing research, writing, summarization, or translation — the editors’ picks for each task (e.g., Claude Opus 4.7 for writing) save hours of manual evaluation.
What can you do with LLM Reference?
- Compare models for a specific task: For example, compare Claude Fable 5 vs GPT‑5.5 to see which scores higher on your required benchmark.
- Track weekly changes: Scan the Pulse page each week to see exactly which models debuted, which prices dropped, and which benchmarks were refreshed — no newsletter, just a filtered log.
- Find the cheapest provider: Every model page shows price per million tokens across multiple providers, so you can ship with the lowest cost for your volume.
Pricing
Free. No registration or API key required to search, compare, or browse the Pulse feed.
FAQ
Is LLM Reference free?
Yes — all features (model directory, side‑by‑side comparisons, editors’ picks, Pulse feed) are fully free with no account required.
How often is the data updated?
New models, price changes, and benchmark scores are added weekly. The front page shows a “this week” summary (e.g., 71 new models, 96 price cuts).
Can I compare models from different providers?
Yes. The compare tool works across any two models in the database, regardless of provider (e.g., Anthropic vs OpenAI vs Google).
