A sourced reference for AI models, developer tools, automation, and deployment. Compare capabilities, pricing units, release status, and evaluation results using links to publisher documentation.
2026-10-09 06:27 UTC · document version 3.62
The review covers the entire matrix: model catalogs and specifications, coding and reasoning comparisons, editors and CLI tools, API providers, browser and desktop automation, infrastructure, guides, and reference sections. Model data lives in the full matrix; dated changes and their sources live in the changelog.
The header records the document review. Historical rows retain their own verification dates, so a current header does not make every older figure current. Release dates, training cutoffs, and verification dates are separate. ⭐ marks exact dates within the review's preceding 30 days.
Start with the task and deployment constraints, then open the relevant model or tool's official source. Compare input and output prices separately, including cache charges, prompt-length thresholds, service tiers, and temporary offers. Subscription allowances and API pricing are different entitlements.
Benchmark scores require the same dataset, version, harness, and tool budget to be comparable. Capability tables and guides provide cross-references; each model or tool has one primary catalog entry. A context limit, parameter count, or leaderboard position alone is not a recommendation.
- Cadence: a full refresh every one to two weeks, plus a same-week entry whenever a major lab ships something. See CHANGELOG.md for the dated history.
- Scope of a refresh: all 94 tracked sections get re-checked in each pass, not only the ones with obvious news. A section that hasn't changed still gets confirmed.
- Sourcing order: official company sites and API docs first, then official GitHub repos and press releases, then model cards from verified Hugging Face orgs, then papers from the model's own authors. Major tech press is a last resort and gets labelled as such.
- What doesn't go in: announced-but-unreleased models, benchmark numbers nobody published, and unofficial model names. Estimates are marked with
~, and missing data stays as-rather than getting filled in with a guess. - Every row carries a verified date so you can see how stale any single entry is instead of trusting one banner at the top.
| Section | Covers |
|---|---|
| Models | Public proprietary APIs, open weights, specialized coding and reasoning models, multimodal generation, specifications, hardware planning, and benchmarks |
| Development tools | Editors, IDE extensions, coding CLIs, provider catalogs, inference services, and GPU clouds |
| Automation | AI browsers, extensions, developer libraries, hosted sandboxes, desktop control, and multi-agent workflows |
| Infrastructure | Embeddings, reranking, video, TTS and transcription, guardrails, RAG, fine-tuning, evaluation, MCP, skills, and routing |
| Guides | Task selection, free-access eligibility, local deployment, subscription costs, and workload-based API versus self-hosting comparisons |
| Reference | Terminology, selected release/price/context comparisons, primary sources, and verification methodology |
Use official company documentation, API rate cards, release repositories, publisher model cards, and benchmark creators' material. Public previews are labeled; restricted programs and planned releases are distinguished from public availability. Missing results stay as -, and estimates state their assumptions.
A reachable link does not verify a claim. A blocked page or timeout does not prove retirement. Before adopting a product, check its current endpoint, license, supported region, account limits, and billing unit in the linked source.
Open an issue or submit a pull request against docs/readme.md, identifying the row and the official source that supports the correction. The contribution guide describes the table formats and categorization rules. Either way, include the primary source. A claim without a link can't be merged.
The documentation is licensed under CC BY 4.0. You may copy, adapt, and redistribute it, including commercially, with attribution to this repository. Individual models, software, and skills have their own licenses.
Maintained by ReadyPixels LLC · Repository