WHAT CHANGED, AND WHEN.

Changelog.

Catalogue editions, method changes, policy, corrections, and site changes — one dated log, also published as an Atom feed at /feed.xml.

Current edition378b9175

Catalogue Sep 11, 2026 · 219 tools · 483 models · 289 verified prices

Subscribe · Atom feed
MethodAIS-V2 · 2026.09-v2

As of Sep 6, 2026 · 10 peer boards: 0 ranked, 4 provisional, 6 building

How we assess
  1. Site

    Redesign shipped: decision-first pages on one design system

    Explore now opens with an edition stamp (tools, models, verified prices, benchmarked models) and one task control; tool rows show a vendor-stated price or n.a. — boilerplate is never shown as data. Profiles lead with a three-cell decision strip (task fit, cost at your workload or vendor price, evidence in this edition) and a peer-comparison sidebar built from Method AIS-V2 groups. Compare starts from legitimate peer pairs. Costs shows a monthly range and median instead of a bare lowest figure, an explicit assumption stack, and a labelled log-scale fit × cost plot with linked hover. Prompts are grouped into seven curated groups with specific/generalized twins paired. Every one of the 800 sitemap URLs kept its address.

  2. Site

    Changelog page and Atom feed now share one source

    This page and /feed.xml are generated from the same dated log at build time, alongside the corrections register and the current edition and method stamps. Editions are hash-pinned; the pointer shown here is the one the live site serves.

  3. Policy

    Commercial policy published; catalogue editions pinned and replayable

    Money can never buy rank: the firewall is now a public page enforced by build tests. Catalogue snapshots ship as hash-pinned immutable editions (pointer next-ba361582) with offline replay, a corrections log, and a grounded catalogue helper. Evidence profiles remain the authority for every claim.

  4. Correction

    Duplicate evidence records in audit coverage — fixed

    The catalogue auditor counted duplicate evidence records once per appearance, overstating canonical-record coverage. Current catalogue edition: the auditor dedupes evidence records by record id before counting coverage.

  5. Catalogue

    Hermes Agent (Nous Research) enters as a provisional entry

    Self-improving, MIT-licensed, multi-model agent with CLI, TUI and Desktop surfaces, persistent memory, skills, subagents, MCP, terminal backends and browser automation. Frontend-specific visual quality is not first-party benchmarked, so it is tracked with no ordinal rank.

  6. Catalogue

    Hermes IDE added as a separate tracked tool

    A different product from Hermes Agent: free, source-available (BSL 1.1) desktop IDE with Agent mode, multi-session and multi-project workflows and first-class git worktrees on macOS, Linux and Windows; v1.3.2 released May 18, 2026.

  7. Catalogue

    Gemini Code Assist row becomes Google Antigravity

    Google replaced the free individual Gemini Code Assist CLI and IDE extensions with Antigravity, an agentic development platform whose IDE lets agents operate the editor, terminal and browser. The row was relabelled; the score was carried over unchanged and is flagged as pending re-review.

  8. Catalogue

    Identity, sharing and catalogue rendering fixed

    Real favicon and icon pack, canonical URL, Open Graph and Twitter cards, JSON-LD, robots.txt, sitemap, Atom feed and a 404 page. Blank cards in the All view fixed. Evidence cards now label themselves from the entity type.

Editions are hash-pinned and immutable; the pointer above is the one the live site serves. Dates are publication dates, not benchmark run dates. Money never buys rank.