Catalogue dump 2026-10-13

Dated dump of the Undominated.ai catalogue as of 2026-10-13.

• 442 model listings · 53 providers · frontier 11
• #1 (quality, standard, rated): google/gemini-4-argon · q 1533.9
• Same allowlisted files as Hugging Face LPH98/undominated-ai-model-pricing (no Artificial Analysis fields)
• LLM catalogue only: image, video and embedding catalogues are not part of this dump.
• Unrated is not zero. Prices and scores are traced to source files on disk / the live site.
• datapackage.json contentHash: sha256:313a5564277c8b8e443b4602f084e11150408d886369ccac1abe77f8885eaebb

Live board: https://undominated.ai/

undominated-check 0.5.0 — guided resource installation

Resource setup now has a guided terminal interface:

npx undominated-check install
# Or start with a specific bundled resource:
npx undominated-check install undominated-evidence-audit

Choose one or more supported tools, project or global scope, and review the destinations before confirming. Existing scripted commands remain available. Cancellation before confirmation writes nothing; existing destinations are refused. MCP setup exports files and registration instructions, with registration kept manual.

The bundle remains 11 skills, 6 agent profiles and 1 MCP server. Other GitHub skills remain available through the Skills CLI.

Verification: 101 package tests; anonymous registry archive equality; 18 fresh resource installations, 69 matching source hashes, eleven fixtures and MCP initialization; real terminal install, decline, Ctrl+C, EOF and dry-run checks. GitHub checks passed on Linux, macOS and Windows. Implementation and checks · Installation guide.

Published npm archive SHA-256:
1e197ca6a9ede5eea19ceaf11f09c47c51ee98b0531c9690e33fc427330f7a24

Catalogue dump 2026-10-06

Dated dump of the Undominated.ai catalogue as of 2026-10-06.

• 447 model listings · 55 providers · frontier 12
• #1 (quality, standard, rated): anthropic/claude-opus-5.5 · q 1511.7
• Same allowlisted files as Hugging Face LPH98/undominated-ai-model-pricing (no Artificial Analysis fields)
• LLM catalogue only: image, video and embedding catalogues are not part of this dump.
• Unrated is not zero. Prices and scores are traced to source files on disk / the live site.
• datapackage.json contentHash: sha256:f0d65940bbc281be497c7d0cd7ea5f44cc2266cdf1f25b462b85a234dfbf9832

Live board: https://undominated.ai/

undominated-check 0.4.0 — verified client installation

Install original resources into a named coding assistant with explicit destinations and verified file hashes.

• install aliases resources install; both spaced flags and --project=... / --target=... forms work.
• Skills target Codex, Claude Code and GitHub Copilot directories. Existing default paths stay compatible with 0.3.1.
• Explicit Claude/GitHub profile targets create native adapters, retain the original licensed files and inherit the host's model, tools and permission policy.
• MCP helpers separate executable/arguments, quote POSIX and PowerShell commands, document project/user scope and leave existing configuration untouched.
• Invalid target combinations, interpolation-sensitive MCP paths, symlinks and existing destinations are refused.
• Optional Bash completion is bundled.

The immutable bundle remains eleven skills, six profiles and one read-only MCP server. All 69 resource source files retain the verified 0.3.1 bytes. The broader source-only library remains a separate distribution.

Validation: 71 CLI tests, 32 MCP tests, 22 public Action tests; a separate 71-case hostile-input installer review; clean archive installation of all 18 resources, 69 hashes and eleven synthetic checks; actual skill discovery in Claude/Codex/Copilot and all six native adapters in Claude/Copilot without inference calls. Linux, macOS and Windows CI passed, including PowerShell argument parsing. Claude/Codex/Cursor MCP handshakes exposed five tools. VS Code registration was verified in an isolated profile; editor execution was not established.

Archive SHA-256: 260ab6679d4068f0e9ea01c6890b09eedad13b5097eb38dc2389c8b784e9a49b.

Installation guide · Evidence and limits

Undominated Check 0.3.1 — installable evidence checks

Installable evidence checks, with corrected boundaries

Undominated Check 0.3.1 bundles 11 original skills, 6 portable agent profiles and the read-only MCP server. Inspect a resource before installing it into an existing project:

npx --yes undominated-check@0.3.1 resources list
npx --yes undominated-check@0.3.1 resources inspect undominated-evidence-audit
npx --yes undominated-check@0.3.1 resources install undominated-evidence-audit --project /absolute/path/to/project --dry-run

Remove --dry-run to install. Node.js 22.12+ is required; skill validators use Python 3.10+ with the standard library. Installation verifies bundled hashes and refuses existing destinations. Profiles are portable instructions that you adapt to your assistant.

Corrected checks

Adverse review of 0.3.0 found unsupported favourable outcomes in five added checkers. This release separates internal-use evidence from redistribution permission, verifies capability preservation in the actual replacement direction, requires an explicit final tier cap, computes large plan totals without decimal-context rounding and compares spread/tier claims exactly.

Validation metadata now says fixture-tested; it does not imply publication status. A successful check validates supplied inputs against its contract, not source truth or universal production suitability.

Release verification

The exact uploaded archive passed 54 source tests, fresh installation of all 18 bundled resources, 69 installed file-hash comparisons, 11 synthetic validators and five MCP tool-definition checks. Anonymous registry archive equality and fresh installation are verified separately before this release is published.

SHA-256 of the attached npm archive:
a4550da5d0e5749b62217a96659f0d8ab91b1ab0a656e1cecd7115a983fe940b

Get started · Resource library · Evidence and corrections

The live website directory contains 71 Skills, 72 Agents and 89 MCP servers as checked 2026-10-07, including third-party resources. Those review counts are separate from this installable original bundle.

Catalogue dump 2026-09-23

Dated dump of the Undominated.ai catalogue as of 2026-09-23.

• 437 model listings · 52 providers · frontier 13
• #1 (quality, standard, rated): anthropic/claude-fable-5.1 · q 1507.6
• Same allowlisted files as Hugging Face LPH98/undominated-ai-model-pricing (no Artificial Analysis fields)
• Unrated is not zero. Prices and scores are traced to source files on disk / the live site.

Live board: https://undominated.ai/

Catalogue 2026-09-22

Dated catalogue dump matching the live board on undominated.ai.

• asOf: 2026-09-22
• models: 425
• providers: 52
• frontier: 12
• #1 (quality, standard): anthropic/claude-fable-5.1 (q 1507.6)

Unrated is not zero. This dump is not a live feed.

Citable catalogue dump 2026-09-19

Licence-gated snapshot of the Undominated.ai catalogue for citation.

• 429 listings, 52 providers, data date 2026-09-19
• LMArena Elo under CC BY 4.0 from the official Hugging Face dataset (attribution in LICENSE.md)
• OpenRouter / vendor list prices, every context-tier rung kept
• No Artificial Analysis field (allowlist in scripts/build-export.mjs in the build tree)
• Combined content hash: sha256:ed75e26e94e2091cd70a611b9342bf27060e13f550532f10b1d5c0e452507361

137 rows carry an Arena score with a confidence interval. 292 rows carry no quality score in this file — unmeasured, or measured and withheld because we may not redistribute the number. null is not zero.

This replaces catalogue-2026-08-31 as the current citable dump. The older dump stays up. Cite this tag, not a screenshot of the live board.

How to cite: CITATION.cff · datasets/README.md

Citable catalogue dump 2026-08-31

Licence-gated snapshot of the Undominated.ai catalogue for citation.

• 405 listings, 50 providers, data date 2026-08-31 (datapackage.json)
• LMArena Elo under CC BY 4.0 from the official Hugging Face dataset (attribution in LICENSE.md)
• OpenRouter / vendor list prices
• No Artificial Analysis field (allowlist in scripts/build-export.mjs)

This is the 2026-Q3 dump. A later quarter should be a new version of the same record, not a second concept DOI, once Zenodo is connected.

Files: models.json, models.csv, README.md, LICENSE.md, datapackage.json.