The trust layer forAI news is missing.
Live rankings, independently verified benchmarks, and honest pricing — distilled from 2,900+ models and 123 trusted feeds into one calm, considered signal.
The whole frontier,
in your pocket.
One live surface for what's moving right now — model shifts, unified scores, price movement and the stories behind them. A preview of the Cadle mobile app, arriving with the beta.
A newsroom's discipline.
A curator's eye.
Models, benchmarks, the Cadle 50, pricing, news and learning — each one sourced and dated, so the question that used to take fourteen tabs takes a single, considered glance.
Model tracking
Frontier, open-weight, image, audio, and multimodal models merged into one ready-to-serve registry.
- OpenAI, Anthropic, Google, Meta, Mistral, xAI, DeepSeek, Alibaba, Cohere, MiniMax, Moonshot, Zhipu and more.
- Context windows, modalities, and provider provenance resolved in-app.
- A background worker combines OpenRouter, Hugging Face, and benchmark-matrix records before the API serves them.
Benchmark tracking
Ten category boards and twelve score columns — importance-weighted Cap percentiles across BenchLM, AA, LLM Stats, Epoch, and HF.
- Unified scores, not a single publisher's self-report.
- Per-model source count so multi-source corroboration is visible.
- Capability, community, value, reliability, and momentum feed the Cadle Cap score.
Cadle 50
Live top-50 models from the unified matrix. Apps, studios, and companies ship as curated shorter lists.
- Models are the live Cadle 50 — always fifty when the matrix is healthy.
- Apps ~15, studios ~12, companies ~12 curated — labeled separately from the model index.
- Podium and movers surface on the home dashboard.
Pricing tracking
Normalized token pricing from OpenRouter, provider connectors, BenchLM, Artificial Analysis, and the model database.
- In/out $/1M normalized into one comparable table.
- 180+ named providers resolved through the lab map.
- Source labels remain attached so users can distinguish provider quotes from benchmark overlays.
News + Learn
123 RSS sources, 280+ live stories, and sixty-plus learning courses — filed, dated, sourced.
- Company blogs, journals, Google News lenses, and quality-web sources across nine desks.
- Learn catalog covers courses and lab paths.
- Nine journals cover AI news, image, video, tools, infrastructure, studios, datasets, innovation, and research.
Signal + dossiers
Lab follow entities, Reddit blend, and per-model dossiers with multi-source provenance.
- OpenAI, Anthropic, DeepMind, Meta, NVIDIA, HF, Mistral, xAI, DeepSeek, Cohere and peers.
- Dossiers pull matrix scores, pricing, and cited sources.
- Tag taxonomy covers ~60 content tags across desks.
Every genre of AI news —
filed, dated, sourced.
A trust-scored AI wire — 123 curated feeds with 280+ live stories across nine desks, auto-tagged and enriched, with Latest, Trending, and For You — then Coverage that opens the original story.
Frontier labs and press desks in one wire — OpenAI, Anthropic, DeepMind, TechCrunch AI, The Verge AI, and peers, scored for trust.
- Company blogs · journals
- Latest · Trending · For You
- Auto-tagged stories
- Coverage → original
Diffusion launches, image APIs, edit pipelines, and studio stills — Midjourney, Flux, Firefly, Ideogram, and research forks.
- Model drops
- Edit tooling
- API pricing shifts
- Entity deep-links
Video foundation models and motion pipelines — Sora-class drops, Kling, Runway Gen, Luma Dream Machine, and research demos.
- Model releases
- Motion tooling
- Latency · cost notes
- Source portfolios
Builder platforms and agent tooling — Cursor, Vercel AI, Replicate, Modal, LangChain, GitHub Copilot, and shipping SDKs.
- Platform launches
- API changes
- Agent frameworks
- Source transparency
Clusters, accelerators, and cloud AI capacity — NVIDIA, AWS ML, Azure AI, CoreWeave, DCD, and The Register.
- Cluster / GPU news
- Cloud AI desks
- Capacity stories
- Enrichment + images
Creative product desks — Runway, Luma, Pika, ElevenLabs, and studio-class launches that ship for makers.
- Studio products
- Creative AI
- Launch coverage
- Model / tool links
Training corpora, eval sets, and data ops — Hugging Face, Databricks AI, LAION-class releases, and annotation stacks.
- HF · Databricks
- Corpus drops
- Eval sets
- Tagged taxonomy
Breakthrough research and invention desks — BAIR, Stanford HAI, The Gradient, Interconnects, The Batch, and lab notes.
- Lab blogs
- Invention tags
- Breakthrough briefs
- For You affinity
Primary papers and technical analysis — arXiv cs.AI/CL/CV/LG, university labs, Epoch AI, and independent researchers.
- Four arXiv feeds
- Lab publications
- Benchmark research
- Original-source links
The signal,
without the newsletter theatre.
Other AI news
Status quo- Scattered across 50+ sources
- Self-reported benchmarks
- Hype-driven headlines
- Stale pricing pages
- No audit trail
- Opinion mixed with data
Cadle
You are here- One registry. 2,900+ models.
- Multi-source importance-weighted Cap
- Data-first, opinion-free
- 2,900+ normalized pricing rows
- Every row sourced and dated
- Cadle 50 + provenance on home
The frontier moves in hours. The information around it moves in threads, hot takes, and screenshots. We think builders deserve a source that moves at the speed of the models — and stays honest doing it.
Built for the gaps
leaderboards leave open.
Honest product statements — not paraphrased social proof or invented upvote counts.
"One Cap board, one Matrix datasheet, and live gateway spreads — without fourteen tabs of conflicting leaderboards."
"Importance-weighted percentiles so ARC-AGI and MMLU never compete raw. Blanks stay blank — never invent zeros."
"OpenRouter, Together, Fireworks, and Groq prices on one desk you can diff against yesterday."
Answers,
up front.
Most boards lean on one publisher's self-report. We unify BenchLM, Artificial Analysis, LLM Stats, Epoch, Hugging Face and more into importance-weighted Cap percentiles — multi-source scores you can cite, not a marketing slide. Price never enters Cap.
123 curated RSS feeds — company blogs, journals, Google News lenses, and quality-web sources — ingest into nine topic desks with 280+ stories on the live desk. Stories are tagged, enriched, and ranked as Latest, Trending, or For You, while Coverage preserves the original link.
It's explainable preference tilting — preferred sources, tags, and affinity on top of momentum — stored locally. Not an opaque ML recommender or cross-device account sync. Source picks and tags are the strong personalization story today.
Cadle combines OpenRouter, direct provider connectors, BenchLM, Artificial Analysis, and stored model prices into one normalized table for input/output cost, speed, latency, and value. Every row keeps its source label, and free tiers are not assigned invented prices.
No. We curate trusted feeds and preserve provenance (source, timestamp, original link). That's a trust layer for AI news — not in-house verification or fact-checking.
Yes. Sign in or create an account and start discovering today. A read-only public API is shipping with the product surface; programmatic rankings and pricing access roll out next.
Discover AI,
without the noise.
The Cadle 50, verified benchmarks, honest pricing, model dossiers, and a news desk worth reading — in one quiet place.